A authentication analysis system and method based on big data
Through dynamic time bending algorithm, isolated forest algorithm, neural network, prefix span algorithm and blockchain technology, the shortcomings of traditional authentication analysis systems in user behavior analysis and credit risk assessment are solved, data security and integrity are improved, and data security and integrity are achieved. More accurate and reliable certification analysis.
Patent Information
- Application Number
- CN202410743293.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-11
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-06-11
AI Technical Summary
Traditional authentication analysis systems are not accurate enough in time series data processing, and it is difficult to capture subtle changes in user behavior, resulting in insufficient security and credit risk assessment, and insufficient data security and integrity.
Dynamic time bending algorithm is used to match and cluster time series shapes, combine isolated forest algorithm to identify abnormal behaviors, neural networks conduct credit risk assessment, use prefix span algorithm to deeply explore user behavior, hidden Markov model to build adaptive authentication model, and Ensure data integrity through blockchain technology.
It improves the accuracy and security of user behavior analysis, enhances the comprehensiveness and accuracy of credit risk assessment, improves the security and integrity of data, and ensures the reliability of certification results and the overall security of the system.
Smart Images

Figure CN118608167B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data authentication, and in particular to an authentication analysis system and method based on big data. Background Art
[0002] Data authentication technology is a field that focuses on using large-scale data sets for identity authentication, transaction verification, or confirmation of content authenticity. The core of this technology field lies in combining advanced data analysis methods such as machine learning, statistical analysis, and pattern recognition to ensure the integrity, security, and credibility of data. This technology is particularly important in areas such as financial services, network security, and health information management, where the authenticity and accuracy of data are crucial.
[0003] Among them, the authentication analysis system based on big data is a system that uses big data technology to improve the efficiency and accuracy of the data authentication process. The main purpose of this system is to verify the authenticity, security and integrity of data by analyzing huge data sets. For example, in the financial industry, this system can be used to prevent fraud and identify abnormal behavior by analyzing transaction patterns; in the field of network security, it can be used to detect and defend against network attacks and identify potential security threats by analyzing network traffic and user behavior.
[0004] In traditional authentication analysis systems, the processing of time series data is not accurate enough, and it is difficult to capture subtle changes in behavior patterns, resulting in insufficient in-depth analysis of user behavior. In terms of security, traditional systems often fail to effectively identify and prevent abnormal security behaviors, posing a major security risk. Credit risk assessment only stays on the surface, lacking in-depth multi-dimensional analysis, making risk assessment results incomplete and inaccurate. The analysis and prediction of user behavior are also often subject to technical limitations, lacking in-depth insights and accurate predictions. The verification of authentication results mostly relies on a single method, lacking an effective cross-validation mechanism, resulting in insufficient reliability of authentication results. In terms of data security and integrity, due to the lack of efficient encryption and chain recording technology, data is vulnerable to damage or tampering, affecting the overall security and reliability of the system. Summary of the invention
[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an authentication analysis system and method based on big data.
[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions: an authentication analysis system based on big data includes a behavior time analysis module, a security anomaly identification module, a credit risk quantification module, a user behavior mining module, an authentication data matching module, an authentication result verification module, and a chain data integrity module;
[0007] The behavior time analysis module uses a dynamic time warping algorithm to match and align the shape of the time series based on the original authentication data set, and applies a sequence clustering method to classify and extract patterns from the time series data, analyze the temporal regularity of user behavior, and generate a temporal behavior feature set;
[0008] The security anomaly identification module uses the isolation forest algorithm to perform anomaly scoring on each instance in the feature set based on the time behavior feature set, identifies abnormal behaviors that differ from the normal pattern, and generates security anomaly indicators;
[0009] The credit risk quantification module uses neural networks to analyze multidimensional risk factors based on security anomaly indicators, classifies and scores risk levels through support vector machines, and conducts a comprehensive quantitative assessment of credit risks to generate a credit risk index;
[0010] The user behavior mining module uses data preprocessing technology to organize and standardize behavior data based on the original authentication data, and applies the prefix span algorithm to deeply mine and analyze user behavior patterns to generate user behavior insights;
[0011] The authentication data matching module uses a hidden Markov model to model the state transition of the data stream based on the credit risk index and user behavior insights, and updates the dynamic Bayesian network according to the model results to match the current data pattern and generate an adaptive authentication model;
[0012] The certification result verification module is based on an adaptive certification model, applies a random forest algorithm to analyze and classify data features, and uses a gradient boosting machine to perform deep learning and analysis of data features, and then cross-validates the certification results through an integrated learning method to generate a comprehensive certification score;
[0013] The chain data integrity module uses blockchain technology to encrypt the authentication data based on the comprehensive authentication score, and stores the encrypted data in the blockchain network for continuous chain recording, thereby generating a chain integrity certificate.
[0014] As a further solution of the present invention, the time behavior feature set includes the time pattern of user behavior, key behavior time points, and statistical characteristics of time series; the security anomaly indicators include abnormal behavior scores, behavior deviations, and frequencies of abnormal behaviors; the credit risk index includes risk levels, key risk factors, and risk trend forecasts; the user behavior insights include user behavior patterns, key behavior characteristics, and behavior change trends; the adaptive authentication model includes network structure adjustment parameters, data relationship weights, and model adaptability indicators; the authentication comprehensive score includes data consistency scores, authentication accuracy, and verification efficiency indicators; the chain integrity proof includes blockchain record information, data integrity check codes, and data consistency proofs.
[0015] As a further solution of the present invention, the behavior time analysis module includes a time dependency analysis submodule, a sequence pattern recognition submodule, and a behavior feature extraction submodule;
[0016] The time dependency analysis submodule uses a dynamic time warping algorithm to perform shape matching and alignment of time series based on the original authentication data set, constructs a distance matrix by calculating the distance between each pair of time points, and uses dynamic programming technology to capture the path with the minimum cumulative distance, performs optimal matching between sequences, and generates a time series alignment set;
[0017] The sequence pattern recognition submodule classifies and extracts patterns from time series data using a sequence clustering method based on a time series alignment set, selects K-means or hierarchical clustering, selects the number of clusters and a distance metric, classifies time series data, identifies patterns and rules, and generates a sequence pattern feature set;
[0018] The behavior feature extraction submodule uses the principal component analysis method to reduce the dimension of the extracted features based on the sequence pattern feature set, extracts the principal components by calculating the covariance matrix of the data, and then reduces the dimension of the data according to the principal components to extract key behavior features and generate a time behavior feature set.
[0019] As a further solution of the present invention, the safety anomaly identification module includes a behavior safety analysis submodule, an abnormal pattern recognition submodule, and a safety index generation submodule;
[0020] The behavior security analysis submodule uses the isolation forest algorithm to score and identify abnormal behaviors based on the time behavior feature set. An isolation tree is constructed by randomly selecting features and selecting the cut value of the features. The number of constructed trees is set to 100, and the number of randomly selected samples for each tree is set to 256. The average path length from the sample point to the leaf is calculated, and samples with shorter path lengths are considered abnormal, thereby generating an abnormal behavior identification set;
[0021] The abnormal pattern recognition submodule uses the K-means clustering algorithm to recognize and classify abnormal patterns based on the abnormal behavior recognition set, sets the number of cluster centers to 5, uses the Euclidean distance as the similarity metric, and optimizes the cluster center positions iteratively to group abnormal behaviors into differentiated categories to generate a patterned abnormal analysis set;
[0022] The safety index generation submodule is based on the patterned anomaly analysis set, uses a scoring system to classify each anomaly pattern into a safety level, sets the safety level range from 1 to 5, where 1 represents low risk, and then generates a safety anomaly index.
[0023] As a further solution of the present invention, the credit risk quantification module includes a risk measurement submodule, a risk dynamic prediction submodule, and an index output submodule;
[0024] The risk measurement submodule uses a neural network to analyze multidimensional risk factors based on security anomaly indicators. The constructed neural network includes two hidden layers, each with 128 neurons. The ReLU activation function is used to adjust the network weights through the gradient descent algorithm. The learning rate is set to 0.01 and the batch size is 32. The nonlinear relationship and intrinsic pattern between risk factors are captured to generate a risk factor analysis set.
[0025] The risk dynamic prediction submodule uses a support vector machine to classify and score risk levels based on a risk factor analysis set. The SVM model selects a radial basis function as a kernel function, the C parameter is set to 1.0 to balance training error and model complexity, and the gamma parameter is set to 0.1 to determine the influence range of the kernel function. The risk factors are classified and processed to generate a risk prediction analysis set.
[0026] The index output submodule is based on the risk prediction analysis set, uses the weighted average method to comprehensively refer to multiple types of risk factors, sets weights for each risk factor, performs weighted calculations based on the degree of impact and importance on credit risk, and then generates a credit risk index.
[0027] As a further solution of the present invention, the user behavior mining module includes a behavior pattern mining submodule, a user habit analysis submodule, and a behavior insight construction submodule;
[0028] The behavior pattern mining submodule uses Python's Pandas library to clean the data based on the original authentication data, and applies the Apriori algorithm to mine association rules, setting the minimum support to 0.01 and the minimum confidence to 0.5. It also uses the mlxtend library to mine frequent itemsets and association rules, analyze user behavior patterns, and generate in-depth analysis of behavior patterns.
[0029] The user habit analysis submodule uses a machine learning algorithm based on in-depth analysis of behavioral patterns, uses the Scikit-learn library to perform K-means cluster analysis, sets the number of clusters to 5, uses the fit_predict method to cluster user habit data, analyzes the behavioral habits of multiple user groups, and uses principal component analysis to reduce data dimensionality, retains the principal components, and generates user habit analysis;
[0030] The behavioral insight construction submodule is based on a comprehensive analysis of user habits. It uses the NumPy and Pandas libraries to merge and reorganize data, combines behavioral patterns and user habit data, and uses the support vector machine algorithm to classify behavior. It uses the SVC function in the Scikit-learn library, sets the kernel function to the radial basis function, and the C parameter to 1.0. The fit method is used to train the model, adjust and optimize the classification results, and generate user behavior insights.
[0031] As a further solution of the present invention, the authentication data matching module includes a network dynamic adjustment submodule, a data relationship matching submodule, and a model updating submodule;
[0032] The network dynamic adjustment submodule uses a hidden Markov model to model the state transition of data streams based on the credit risk index and user behavior insights. The number of states in the model is set to 5, and the transition probability is dynamically adjusted according to the behavior frequency in the data stream, and the state change trend in the data stream is analyzed to generate a state transition analysis.
[0033] The data relationship matching submodule is based on state transition analysis, adjusts the network structure according to the results of the hidden Markov model, resets the weights of nodes and edges, matches the relationship pattern between data, optimizes the network structure using Bayesian parameter estimation, and generates data relationship matching analysis;
[0034] The model update submodule uses the gradient boosting tree algorithm to adjust the model based on data relationship matching analysis and state transition analysis, sets the learning rate to 0.05 and the maximum depth to 4, and reflects the current data flow and behavior trend of the model through parameter adjustment to generate an adaptive authentication model.
[0035] As a further solution of the present invention, the authentication result verification module includes a learning strategy synthesis submodule, a result verification submodule, and a comprehensive score generation submodule;
[0036] The learning strategy synthesis submodule is based on the adaptive authentication model and uses the random forest algorithm to analyze and classify data features. The number of trees in the random forest algorithm is set to 100, the maximum depth is 10, and the Gini impurity is used as the splitting criterion to perform feature importance evaluation and classification decisions, analyze data features and identify key classification criteria, and generate feature analysis results;
[0037] The result verification submodule uses a gradient boosting machine to perform deep learning and analysis on data features based on the feature analysis results. The gradient boosting machine sets the learning rate to 0.1, the maximum depth to 5, and the loss function to deviance. It refines data features and identifies deep data patterns through iterative optimization, thereby generating deep feature analysis results.
[0038] The comprehensive score generation submodule is based on the results of deep feature analysis, adopts an integrated learning method, combines multiple differentiation models, integrates the random forest, support vector machine and gradient boosting decision tree algorithms of the Scikit-Learn library, sets the weights and parameters of multiple models, uses the cross_val_score function for cross-validation, uses a soft voting mechanism, combines the probability estimates of multiple model outputs, calculates the average probability score, and generates a comprehensive certification score.
[0039] As a further solution of the present invention, the chained data integrity module includes a blockchain construction submodule, a data invariance assurance submodule, and an integrity proof generation submodule;
[0040] The blockchain construction submodule is based on the comprehensive certification score, uses blockchain technology, and creates a blockchain structure through a programming language. Each block uses the SHA-256 hash algorithm to generate a unique hash value, processes the data record, the hash value of the previous block, and the current data into a string, applies the hash algorithm to convert it into a unique hash value of a fixed length, and continuously concatenates each block to generate an encrypted data chain;
[0041] The data invariance guarantee submodule is based on the encrypted data chain, and uses a combination of hash functions and timestamps through programming to lock the content of each block, apply a hash function to the data and timestamp in each block to generate the current hash value, assist in data recording in the blockchain network, and generate data invariance records;
[0042] The integrity proof generation submodule uses a blockchain integrity verification program based on data invariance records to check the data and structure of each block in the blockchain, uses programming logic to verify the integrity and change item status of each block in the chain, and generates a chain integrity proof.
[0043] A big data-based authentication analysis method, which is performed based on the big data-based authentication analysis system, comprises the following steps:
[0044] S1: Based on the original authentication dataset, a dynamic time warping algorithm is used to perform shape matching and alignment of time series, and a sequence clustering method is used to classify and extract patterns from time series data, analyze the temporal regularity of user behavior, extract key temporal features, and generate a temporal behavior feature set;
[0045] S2: Based on the temporal behavior feature set, an isolation forest algorithm is used to perform anomaly scoring on each instance, and a security anomaly assessment is generated by analyzing and identifying abnormal behaviors that differ from regular user behavior patterns;
[0046] S3: Based on the security anomaly assessment, a neural network is used to analyze multidimensional risk factors, and risk levels are classified and scored by a support vector machine to conduct a comprehensive quantitative assessment of credit risk and generate a credit risk index;
[0047] S4: Based on the original authentication data, the data preprocessing technology is used to organize and standardize the behavior data, and the prefix span algorithm is applied to deeply mine and analyze the user behavior patterns to generate user behavior insight results;
[0048] S5: Based on the credit risk index and user behavior insight results, a hidden Markov model is used to model the state transition of the data stream, and a dynamic Bayesian network is updated according to the model results to match the current data pattern, thereby generating an adaptive authentication model;
[0049] S6: Based on the adaptive authentication model, random forest, gradient boosting machine and ensemble learning methods are used to analyze and verify data features, and then blockchain technology is used to encrypt and process the authentication data to generate a chain integrity certificate.
[0050] Compared with the prior art, the advantages and positive effects of the present invention are:
[0051] In the present invention, the dynamic time warping algorithm realizes the precise matching and alignment of time series in the behavior time analysis, providing more accurate time dimension insights for user behavior analysis. The isolation forest algorithm effectively identifies abnormal behaviors in security anomaly identification and enhances the security of the system. The use of neural networks and support vector machines for credit risk assessment realizes the multi-dimensional in-depth analysis of risk factors and improves the comprehensiveness and accuracy of credit risk assessment. The prefix span algorithm in user behavior mining deeply analyzes user behavior patterns and enhances the understanding and prediction capabilities of user behavior. The construction and application of adaptive authentication models enable the system to flexibly match different data patterns and improve the adaptability and flexibility of data processing. The cross-validation method of authentication results improves the reliability and accuracy of authentication results. The application of blockchain technology ensures the security and integrity of data and provides a powerful data protection mechanism for the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 is a system flow chart of the present invention;
[0053] Figure 2 It is a schematic diagram of the system framework of the present invention;
[0054] Figure 3 It is a flow chart of the behavior time analysis module of the present invention;
[0055] Figure 4 This is a flow chart of the security anomaly identification module of the present invention;
[0056] Figure 5 This is a flow chart of the credit risk quantification module of the present invention;
[0057] Figure 6 This is a flow chart of the user behavior mining module of the present invention;
[0058] Figure 7 This is a flow chart of the authentication data matching module of the present invention;
[0059] Figure 8 It is a flow chart of the authentication result verification module of the present invention;
[0060] Fig. 9 It is a flow chart of the chain data integrity module of the present invention;
[0061] Fig.10 It is a schematic diagram of the method steps of the present invention. DETAILED DESCRIPTION
[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0063] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.
[0064] Embodiment 1
[0065] See also Figure 1 to Figure 2 ,The present invention provides a technical solution: a big data-based authentication analysis system includes a behavior time analysis module, a security anomaly identification module, a credit risk quantification module, a user behavior mining module, an authentication data matching module, an authentication result verification module, and a chain data integrity module;
[0066] The behavior time analysis module uses a dynamic time warping algorithm to match and align the shape of time series based on the original authentication data set. It also applies a sequence clustering method to classify and extract patterns from time series data, analyze the temporal regularity of user behavior, and generate a temporal behavior feature set.
[0067] The security anomaly identification module uses the isolation forest algorithm to score each instance in the feature set based on the temporal behavior feature set, identify abnormal behaviors that differ from normal patterns, and generate security anomaly indicators;
[0068] The credit risk quantification module uses neural networks to analyze multi-dimensional risk factors based on security anomaly indicators, classifies and scores risk levels through support vector machines, and conducts a comprehensive quantitative assessment of credit risks to generate a credit risk index;
[0069] The user behavior mining module uses data preprocessing technology to organize and standardize behavior data based on the original authentication data, and applies the prefix span algorithm to deeply mine and analyze user behavior patterns to generate user behavior insights;
[0070] The authentication data matching module uses hidden Markov models to model the state transition of data streams based on credit risk index and user behavior insights. It also updates the dynamic Bayesian network based on the model results to match the current data pattern and generate an adaptive authentication model.
[0071] The certification result verification module is based on the adaptive certification model, and uses the random forest algorithm to analyze and classify data features. It also uses the gradient boosting machine to perform deep learning and analysis of data features, and then uses the integrated learning method to cross-validate the certification results to generate a comprehensive certification score.
[0072] The chain data integrity module is based on the comprehensive authentication score and uses blockchain technology to encrypt the authentication data. At the same time, the encrypted data is stored in the blockchain network for continuous chain recording, thereby generating a chain integrity certificate.
[0073] The temporal behavior feature set includes the temporal patterns of user behavior, key behavior time points, and statistical characteristics of time series. Security anomaly indicators include abnormal behavior scores, behavior deviations, and frequencies of abnormal behaviors. Credit risk indexes include risk levels, key risk factors, and risk trend forecasts. User behavior insights include user behavior patterns, key behavior characteristics, and behavior change trends. The adaptive authentication model includes network structure adjustment parameters, data relationship weights, and model adaptability indicators. The comprehensive authentication score includes data consistency scores, authentication accuracy, and verification efficiency indicators. The chain integrity proof includes blockchain record information, data integrity check codes, and data consistency proofs.
[0074] In the behavior time analysis module, the dynamic time warping algorithm is used to achieve shape matching and alignment of time series through the original authentication data set. Specifically, this algorithm dynamically constructs a path by calculating the minimum distance between each time point to achieve the best match between two time series. This process involves the horizontal and vertical coordinate values of each point in the time series, and the points on the time axis are adjusted to minimize the overall path distance. Next, the sequence clustering method is applied to classify and extract patterns from the time series data. Sequence clustering calculates the similarity between time series and classifies sequences with similar patterns into a group, thereby identifying the temporal regularity of user behavior. This method covers the extraction of time series features, including but not limited to frequency, amplitude, and duration, as well as cluster analysis based on these features. Through these steps, a temporal behavior feature set is generated, which records the temporal characteristics of user behavior in detail and provides a basis for data analysis in subsequent modules.
[0075] In the security anomaly identification module, based on the temporal behavior feature set, the isolation forest algorithm is used to score the anomaly of each instance in the feature set. The isolation forest algorithm recursively builds a binary tree structure by randomly selecting a feature and a split value of the feature until all points are isolated. In this process, normal data points require more steps to be isolated, while abnormal data points are isolated earlier. The algorithm calculates the average path length required for each data point to be isolated and uses this as a basis to evaluate its degree of anomaly. In this way, the module can effectively identify abnormal behaviors that differ from normal patterns and generate security anomaly indicators. These indicators provide an important basis for identifying potential security threats and enhance the security of the system.
[0076] In the credit risk quantification module, neural networks are used to analyze multidimensional risk factors based on security anomaly indicators, and then support vector machines (SVMs) are used to classify and score risk levels. Neural networks are used here to process and analyze risk factors from different sources, such as user behavior patterns, historical credit records, etc., to extract deep features and associations. Then, SVMs are applied to these features to divide data points into different risk levels by constructing one or more hyperplanes. The core of SVM is to find the optimal classification boundary and maximize the margin between different categories. The process includes selecting a suitable kernel function and adjusting penalty parameters to optimize the classification effect. Finally, the module comprehensively evaluates credit risk and generates a credit risk index, which provides a quantitative basis for subsequent credit decisions and enhances the accuracy and efficiency of risk management.
[0077] In the user behavior mining module, based on the original authentication data, the behavior data is first organized and standardized using data preprocessing techniques, including cleaning, normalization, and encoding steps to ensure data quality. Subsequently, the prefix span algorithm is applied to deeply mine and analyze user behavior patterns. The prefix span algorithm reveals the inherent laws of user behavior by identifying frequently occurring behavioral sequence patterns. This process involves setting a minimum support threshold to identify frequent item sets and using a prefix tree structure to efficiently store and retrieve these patterns. In this way, the module is able to generate user behavior insights that reflect users' typical behavior patterns and preferences, providing data support for personalized services and user experience optimization.
[0078] In the authentication data matching module, the state transition of the data stream is modeled using a hidden Markov model based on the credit risk index and user behavior insights. The model predicts the estimated state transitions in the user behavior sequence by calculating the transition probabilities between states. Then, based on the output of the model, the Bayesian network is dynamically updated to reflect the latest user behavior and risk patterns. The Bayesian network is used here to integrate evidence from different sources and calculate the posterior probabilities of various hypotheses. Through these steps, the module is able to match the current data pattern and generate an adaptive authentication model. This model is able to dynamically adapt to changes in user behavior and improve the accuracy and efficiency of the authentication process.
[0079] In the certification result verification module, based on the adaptive certification model, the random forest algorithm is first applied to analyze and classify the data features. Random forest improves the accuracy and robustness of classification by constructing multiple decision trees and combining their prediction results. Each tree uses a randomly selected feature subset during training, which can reduce the risk of overfitting of the model. Next, the gradient boosting machine is used to perform deep learning and analysis of data features to further improve the prediction ability of the model. The gradient boosting machine optimizes the overall classification performance by gradually building weak learners and focusing on samples that were misclassified in the previous round. Finally, the certification results are cross-validated through an ensemble learning method to ensure the reliability and accuracy of the certification results. Through these steps, the module generates a comprehensive certification score, which reflects the credibility of the certification results and is crucial to maintaining system security and user trust.
[0080] In the chained data integrity module, blockchain technology is used to encrypt the authentication data based on the comprehensive authentication score. Blockchain technology ensures the immutability and transparency of data through distributed ledgers and encryption technology. After encryption, the authentication data is distributedly stored in the blockchain network to form a continuous chain record. This method not only increases the security of data storage, but also provides a means of verifying data integrity. Each data block contains the hash value of the previous block, as well as its own timestamp and data content. This design ensures that any attempt to tamper with the stored data will be quickly discovered. Through these steps, the module generates a chained integrity proof, which is a strong guarantee for data security and integrity and is essential for maintaining long-term data trust and compliance.
[0081] See also Figure 3 ,The behavior time analysis module includes a time dependency analysis submodule, a sequence pattern recognition submodule, and a behavior feature extraction submodule;
[0082] The time dependency analysis submodule uses a dynamic time warping algorithm to perform shape matching and alignment of time series based on the original authentication data set. It calculates the distance between each pair of time points to construct a distance matrix. It also uses dynamic programming technology to capture the path with the minimum cumulative distance, perform optimal matching between sequences, and generate a time series alignment set.
[0083] The sequence pattern recognition submodule uses the sequence clustering method to classify and extract patterns from time series data based on the time series alignment set. It selects K-means or hierarchical clustering, selects the number of clusters and distance metrics, classifies time series data, identifies patterns and rules, and generates a sequence pattern feature set.
[0084] The behavior feature extraction submodule uses the principal component analysis method to reduce the dimension of the extracted features based on the sequence pattern feature set. It extracts the principal components by calculating the covariance matrix of the data, and then reduces the dimension of the data according to the principal components to extract the key behavior features and generate a temporal behavior feature set.
[0085] In the time dependency analysis submodule, the dynamic time warping algorithm is used to match and align the shape of the time series through the original authentication data set, ensuring accurate comparison and analysis of the time series data. The dynamic time warping algorithm first calculates the distance between each pair of time points and constructs a distance matrix. Each element of this matrix represents the distance between the corresponding time points in the two time series. For example, if there are two time series with lengths of M and N, the algorithm will construct an M×N matrix, in which each element reflect[i][j] represents the distance between the i-th point in the first sequence and the j-th point in the second sequence. Subsequently, the dynamic programming technique is used to start from the upper left corner of the distance matrix and gradually move to the lower right corner, and each move selects the path with the minimum cumulative distance. Specifically, in the dynamic programming process, for each point reflect[i][j] in the matrix, the cumulative distance with its left reflect[i][j-1], the upper reflect[i-1][j], and the upper left reflect[i-1][j-1] is calculated, and the minimum value is selected as the cumulative distance of the current point. In this way, when we reach the lower right corner of the matrix, we have found the minimum cumulative distance between the two time series, that is, the optimal matching path. Finally, we generate a time series alignment set based on this path. This set records the matching of the two time series in detail, providing accurate basic data for subsequent pattern recognition.
[0086] In the sequence pattern recognition submodule, based on the time series alignment set, the sequence clustering method is applied to classify and extract patterns from the time series data. K-means or hierarchical clustering is selected as the clustering method. For K-means clustering, the number of clusters K is first determined, and the optimal K value is estimated by the elbow rule or the silhouette coefficient. Then, K time series are randomly selected as the initial cluster centers, and each time series is assigned to the nearest cluster center according to the distance metric between time series (such as Euclidean distance or DTW distance). Next, the center of each cluster is updated to the average value of all sequences in the cluster, and the assignment and update process is repeated until the clustering result is stable or the predetermined number of iterations is reached. For hierarchical clustering, each time series starts as a separate cluster, and the closest clusters are gradually merged until the predetermined number of clusters is reached or a certain distance threshold is reached. Regardless of the method used, the clustering process can identify common patterns and regularities in the time series and generate a sequence pattern feature set. This feature set records the time series patterns of different categories in detail, providing key information for further behavioral analysis.
[0087] In the behavioral feature extraction submodule, the principal component analysis (PCA) method is used to reduce the dimensionality of the extracted features based on the sequence pattern feature set. PCA first calculates the covariance matrix of the data to quantify the degree of association between different features. For example, if there are N features, an N×N covariance matrix is constructed, in which each element represents the covariance between two features. Then, the principal components are extracted by solving the eigenvalues and eigenvectors of the covariance matrix. In this process, the eigenvectors with larger eigenvalues represent the main variation direction in the data, while the eigenvectors with smaller eigenvalues can be ignored to reduce the dimension of the data. The first few principal components are selected as the new feature set, which retains most of the information of the original data but has a lower dimension. Finally, the key behavioral features are extracted based on these principal components, and a temporal behavioral feature set is generated. This feature set simplifies the original features, retains the most important information, and provides effective data support for subsequent analysis and decision-making.
[0088] Assume that in an authentication analysis system based on big data, the original authentication data set contains the login behavior data of users at different time points. For example, the data items include login time, duration, login location, etc. Each data item has a corresponding simulated value, such as login time is 08:00, duration is 30 minutes, and login location is "New York". In the time dependency analysis submodule, these data are first matched and aligned by the dynamic time warping algorithm to generate a time series alignment set, which shows the similarities and differences between the login patterns of different users. In the sequence pattern recognition submodule, these aligned time series are clustered by K-means to identify similar login behavior patterns, such as morning log-ins, night log-ins, etc. In the behavior feature extraction submodule, PCA dimensionality reduction is used to extract key features from complex login behavior data, such as the most common login time and duration, to generate a time behavior feature set. These feature sets can be used to further analyze user behavior patterns and assist in authentication decisions. For example, if a user suddenly changes his or her regular login pattern, this is a potential security risk signal that requires further review and verification.
[0089] See also Figure 4 ,The security anomaly identification module includes a behavioral security analysis submodule, an omaly pattern recognition submodule, and a security indicator generation submodule;
[0090] The behavior security analysis submodule uses the isolation forest algorithm to score and identify abnormal behaviors based on the temporal behavior feature set. It constructs an isolation tree by randomly selecting features and selecting the cutoff value of the features. The number of trees to be constructed is set to 100, and the number of randomly selected samples for each tree is set to 256. The average path length from the sample point to the leaf is calculated, and samples with shorter path lengths are considered abnormal, thereby generating an abnormal behavior identification set.
[0091] The abnormal pattern recognition submodule uses the K-means clustering algorithm to recognize and classify abnormal patterns based on the abnormal behavior recognition set. The number of cluster centers is set to 5, and the Euclidean distance is used as the similarity metric. The abnormal behaviors are grouped into differentiated categories by iteratively optimizing the cluster center positions to generate a patterned abnormal analysis set.
[0092] The security indicator generation submodule is based on the patterned anomaly analysis set. It uses a scoring system to classify the security level of each anomaly pattern, setting the security level range from 1 to 5, where 1 represents low risk, and then generates security anomaly indicators.
[0093] In the behavioral security analysis submodule, the isolation forest algorithm is used to score and identify abnormal behaviors through the temporal behavior feature set. The isolation forest algorithm is particularly suitable for processing high-dimensional data and can effectively identify rare or abnormal behavior patterns. First, the algorithm randomly selects features in the feature set and selects the cut value of the feature to construct an isolation tree. In actual operation, the number of trees constructed is set to 100, ensuring sufficient sample coverage and result accuracy. During the construction of each tree, the number of randomly selected samples is set to 256, which is the result of balancing computational efficiency and sample representativeness. For each isolated tree, the average path length from the sample point to the leaf is calculated. In the isolation forest, samples with shorter path lengths are considered abnormal because they are more likely to be isolated. This process is performed by comparing the path length of each sample point with the path length of randomly generated samples. If the path length of a sample point is significantly lower than the average path length of the random sample, it is marked as abnormal. After this series of operations, an abnormal behavior identification set is generated, which records in detail the behavior patterns identified as abnormal, providing key data for subsequent analysis and response.
[0094] In the abnormal pattern recognition submodule, based on the abnormal behavior recognition set, the K-means clustering algorithm is applied to identify and classify abnormal patterns. K-means clustering is a widely used clustering technique that can effectively group data into different categories. First, the number of cluster centers is set to 5, which is determined based on the distribution of the data and the types of expected abnormal patterns. Euclidean distance is used as the similarity metric because it is a common and intuitive method for calculating the distance between different data points. In the clustering process, 5 cluster centers are first randomly initialized, and then each abnormal behavior is assigned to the nearest cluster center. Subsequently, the position of the cluster center is updated by calculating the mean of all points in each cluster, and this assignment and update process is repeated until the position of the cluster center is stable or the predetermined number of iterations is reached. This process ensures the accuracy and robustness of the final clustering result by iteratively optimizing the position of the cluster center. Finally, the abnormal behaviors are grouped into differentiated categories to generate a patterned abnormal analysis set. This set records the abnormal patterns of different categories in detail, providing a basis for identifying specific types of security threats.
[0095] In the security indicator generation submodule, a scoring system is used to classify the security level of each abnormal pattern based on the patterned anomaly analysis set. This process involves quantitatively evaluating the potential risks of different abnormal patterns. First, the security level is set from 1 to 5, where 1 represents low risk and 5 represents high risk. This rating system is customized based on factors such as the frequency, severity, and potential impact of abnormal behavior. The scoring system determines the corresponding security level of each abnormal pattern in the patterned anomaly analysis set based on its characteristics, such as the duration, frequency, and number of users involved. For example, an abnormal behavior that occurs frequently and affects multiple users is assigned a higher security level. Through these detailed evaluations and classifications, security anomaly indicators are finally generated, which provide system administrators with a clear and quantitative view of the current security status of the system, helping them identify and prioritize the most urgent and serious security threats.
[0096] Assume that in a big data-based authentication analysis system, the temporal behavior feature set contains the login and operation behavior data of different users within a week, such as login time, operation frequency, and device type used. For example, one user logs in using an office computer at 8 am every weekday, while another user logs in using a mobile phone late at night on weekends. In the behavioral security analysis submodule, these data are processed by the isolation forest algorithm to identify abnormal behavior patterns such as logging in using uncommon devices late at night on weekends. In the abnormal pattern recognition submodule, these abnormal behaviors are classified into 5 different categories by the K-means clustering algorithm, such as time-based anomalies, device type-based anomalies, etc. In the security indicator generation submodule, these abnormal patterns are rated as different security levels according to their potential risks, and detailed security anomaly indicators are generated. These indicators can help system administrators quickly identify and respond to potential security threats, such as conducting more in-depth investigations or taking preventive measures for specific types of abnormal patterns.
[0097] See also Figure 5 ,The credit risk quantification module includes the risk measurement submodule, the risk dynamic prediction submodule, and the index output submodule;
[0098] The risk measurement submodule uses a neural network to analyze multidimensional risk factors based on security anomaly indicators. The constructed neural network includes two hidden layers, each with 128 neurons. The ReLU activation function is used, and the network weights are adjusted by the gradient descent algorithm. The learning rate is set to 0.01 and the batch size is 32. The nonlinear relationship and intrinsic pattern between risk factors are captured to generate a risk factor analysis set.
[0099] The risk dynamic prediction submodule uses support vector machines to classify and score risk levels based on the risk factor analysis set. The SVM model selects the radial basis function as the kernel function, the C parameter is set to 1.0 to balance the training error and model complexity, and the gamma parameter is set to 0.1 to determine the influence range of the kernel function. The risk factors are classified and processed to generate a risk prediction analysis set.
[0100] The index output submodule is based on the risk prediction analysis set and uses the weighted average method to comprehensively refer to multiple types of risk factors, set weights for each risk factor, and perform weighted calculations based on the degree of impact and importance on credit risk to generate a credit risk index.
[0101] In the risk measurement submodule, a neural network is used to analyze multidimensional risk factors through security anomaly indicators. The constructed neural network includes two hidden layers, each with 128 neurons, to fully capture the relationship between complex risk factors. The ReLU (Rectified Linear Unit) activation function is used, which is one of the commonly used activation functions in deep learning because it can effectively alleviate the gradient vanishing problem. During the training of the neural network, the network weights are adjusted by the gradient descent algorithm to minimize the prediction error. The learning rate is set to 0.01, which is neither too large to cause oscillations in the learning process nor too small to cause the learning speed to be too slow. In addition, the batch size is set to 32, which can strike a balance between computational efficiency and memory limitations. The neural network inputs multidimensional risk factors, such as user behavior data, historical anomaly records, etc., and uses multi-layer nonlinear transformations to capture the complex relationships and inherent patterns between these factors. Such processing can reveal the deep connection between risk factors, thereby generating a risk factor analysis set. This analysis set provides the system with deeper risk insights and helps to more accurately assess and manage potential risks.
[0102] In the risk dynamic prediction submodule, support vector machine (SVM) is used to classify and score risk levels based on the risk factor analysis set. SVM is an effective classification tool, especially suitable for high-dimensional data. The radial basis function (RBF) is selected as the kernel function because it can handle the interaction of nonlinear features and is one of the commonly used kernel functions in SVM. The C parameter is set to 1.0 to balance the training error and model complexity, avoiding overfitting while ensuring that the model has enough flexibility to capture the characteristics of the data. The gamma parameter is set to 0.1 to determine the influence range of the kernel function. This value affects the distribution of data mapped to the new feature space, thereby affecting the accuracy of classification. By classifying risk factors, SVM can effectively divide risk factors into different levels. The final risk prediction analysis set records the predicted classification of each risk factor in detail, which provides an important basis for dynamic management and early warning of risks.
[0103] In the index output submodule, based on the risk prediction analysis set, the weighted average method is used to comprehensively refer to multiple types of risk factors. Weights are set for each risk factor, and these weights are determined according to the degree of impact and importance of the risk factor on credit risk. For example, higher weights are given to risk factors with greater impact, such as frequent abnormal login behaviors. The weighted calculation process involves multiplying the score of each risk factor by its corresponding weight, and then summing them up to obtain a comprehensive risk score. This method can comprehensively consider various risk factors and assess credit risk more comprehensively. The credit risk index finally generated is a quantitative assessment of the overall credit risk of the user or system. This index provides strong data support for credit management and decision-making, helping to identify high-risk individuals or behaviors, so as to take corresponding risk control measures.
[0104] Assume that in an authentication analysis system based on big data, the security anomaly index includes information such as the number of user logins, login time, and operation frequency. For example, one user logs in multiple times every day late at night within a week, and another user frequently tries to log in with different passwords in a short period of time. These data are input into the neural network. In the risk measurement submodule, the network analyzes these data to reveal the relationship between user behavior and risk level and generate a risk factor analysis set. In the risk dynamic prediction submodule, SVM classifies the user's risk level based on these analysis results. For example, frequent late-night logins are classified as higher risk. In the index output submodule, a comprehensive credit risk index is calculated based on these classification results and the set weights. For example, users who frequently log in late at night get a higher risk index. This credit risk index can be used to evaluate the user's overall credit status and provide an important quantitative basis for credit management and decision-making.
[0105] See also Figure 6 ,The user behavior mining module includes the behavior pattern mining submodule, the user habit analysis submodule, and the behavior insight building submodule;
[0106] The behavior pattern mining submodule uses Python's Pandas library to clean the data based on the original authentication data, and applies the Apriori algorithm to mine association rules. It sets the minimum support to 0.01 and the minimum confidence to 0.5, and uses the mlxtend library to mine frequent itemsets and association rules, analyze user behavior patterns, and generate in-depth analysis of behavior patterns.
[0107] The user habit analysis submodule uses machine learning algorithms based on in-depth analysis of behavioral patterns. It uses the Scikit-learn library to perform K-means clustering analysis, sets the number of clusters to 5, and uses the fit_predict method to cluster user habit data to analyze the behavioral habits of multiple user groups. It also uses principal component analysis to reduce data dimensionality, retains the principal components, and generates user habit analysis.
[0108] The behavior insight construction submodule is based on comprehensive analysis of user habits. It uses NumPy and Pandas libraries to merge and reorganize data, combines behavior patterns and user habit data, and applies the support vector machine algorithm to classify behavior. It uses the SVC function in the Scikit-learn library, sets the kernel function to the radial basis function, and the C parameter to 1.0. The fit method is used to train the model, adjust and optimize the classification results, and generate user behavior insights.
[0109] In the behavior pattern mining submodule, the original authentication data is cleaned and processed by Python's Pandas library, and then the Apriori algorithm is applied to perform association rule mining and generate a deep analysis of the behavior pattern. The data cleaning process first includes removing missing values, outliers, and duplicates to ensure the accuracy and consistency of the data. For example, for user login data, records with incomplete timestamps or abnormal login durations need to be removed. Before the Apriori algorithm is applied, the data needs to be converted into a format suitable for association rule mining, such as converting time series data into transaction records. The minimum support is set to 0.01 and the minimum confidence is set to 0.5. These parameters determine the importance thresholds of item sets and rules during the mining process. The mlxtend library is used to perform frequent item set and association rule mining. For example, by calculating the combined frequency of various user behaviors (such as login, browsing, purchase, etc.), common behavior patterns are found. Through this series of processing, the behavior pattern mining submodule can analyze user behavior patterns and generate a deep analysis of behavior patterns. This analysis can reveal the potential correlation between user behaviors and provide a basis for subsequent user behavior prediction and personalized recommendations.
[0110] In the user habit analysis submodule, based on the in-depth analysis of behavioral patterns, machine learning algorithms are used to cluster and analyze user habits. The Scikit-learn library is used for K-means clustering analysis, and the number of clusters is set to 5, which means that users are divided into 5 different groups based on their behavioral patterns. The fit_predict method is used to cluster the user habit data, which first estimates the cluster center and then assigns each user to the nearest cluster based on the nearest neighbor principle. For example, if a user's login and purchase behavior are very close to a cluster center, the user will be assigned to this cluster. At the same time, principal component analysis (PCA) is used to reduce the data dimension and retain the most important principal components. Dimensionality reduction is to improve the efficiency and accuracy of clustering, and only those features that have the greatest impact on the clustering results are retained. The final generated comprehensive analysis of user habits can reveal the behavioral habits of different user groups, which is crucial for understanding user behavior, improving user experience, and improving service quality.
[0111] In the behavior insight building submodule, based on the comprehensive analysis of user habits, the NumPy and Pandas libraries are used to merge and reorganize the data, combining the behavior patterns and user habit data. The support vector machine (SVM) algorithm is used for behavior classification, using the SVC function in the Scikit-learn library. The kernel function is set to the radial basis function (RBF) because it can effectively handle nonlinear problems. The C parameter is set to 1.0, which determines the model's tolerance for misclassification. The model is trained using the fit method, and the classification results are adjusted and optimized during the process to improve the accuracy and generalization ability of the model. For example, by adjusting the C parameter and the gamma parameter, the complexity of the model and its sensitivity to data features can be controlled. Finally, user behavior insights are generated, which can not only provide in-depth analysis of user behavior, but also help formulate more accurate marketing strategies and user services.
[0112] In the authentication analysis system based on big data, the original authentication data includes the user's login time, operation type, usage time, etc. For example, one user frequently logs in and operates from 9 to 11 am on weekdays, and another user browses for a long time on weekend afternoons. In the behavior pattern mining submodule, these data are cleaned and analyzed by the Apriori algorithm, and behavioral patterns such as "users who log in in the morning on weekdays tend to perform short-term operations" are discovered. In the user habit analysis submodule, these behavior patterns are used for K-means clustering to divide users into different groups based on their behavior patterns, and PCA dimensionality reduction is used to improve clustering efficiency. In the behavior insight construction submodule, these behavior patterns are classified by the SVM algorithm, such as distinguishing high-frequency operation users and long-time browsing users, thereby generating in-depth user behavior insights, which help understand user behavior patterns and provide support for the formulation of personalized service strategies.
[0113] See also Figure 7 ,The authentication data matching module includes a network dynamic adjustment submodule, a data relationship matching submodule, and a model updating submodule;
[0114] The network dynamic adjustment submodule uses the hidden Markov model to model the state transition of data streams based on the credit risk index and user behavior insights. The number of states in the model is set to 5, and the transition probability is dynamically adjusted according to the behavior frequency in the data stream. The state change trend in the data stream is analyzed to generate state transition analysis.
[0115] The data relationship matching submodule is based on state transition analysis, adjusts the network structure according to the results of the hidden Markov model, resets the weights of nodes and edges, matches the relationship patterns between data, uses Bayesian parameter estimation to optimize the network structure, and generates data relationship matching analysis;
[0116] The model update submodule adjusts the model based on data relationship matching analysis and state transition analysis, and applies the gradient boosting tree algorithm. It sets the learning rate to 0.05 and the maximum depth to 4. Through parameter adjustment, it reflects the current data flow and behavior trend of the model and generates an adaptive authentication model.
[0117] In the network dynamic adjustment submodule, based on the credit risk index and user behavior insights, the hidden Markov model is used to model the state transition of the data stream. The hidden Markov model is a statistical model used to describe an observable sequence randomly generated by an unobservable hidden Markov chain. In this model, the number of states is set to 5, which means that the model divides the data stream into 5 different hidden states. Each state corresponds to a specific pattern of user behavior or credit risk. The key to the model is to determine the state transition probability matrix and the observation probability matrix. The state transition probability matrix describes the probability of transitioning from one state to another, while the observation probability matrix describes the probability of observing a specific behavior in a given state. These probabilities are dynamically adjusted based on the frequency of behaviors in the data stream to reflect the changing trend of the actual data. For example, if a user behavior becomes more frequent in the data, the model will adjust the probability of this behavior being associated with a specific state accordingly. In this way, the hidden Markov model is able to analyze the state change trend in the data stream and generate state transition analysis. This analysis reveals the dynamic relationship between user behavior and credit risk status, which is crucial for understanding user behavior patterns and predicting future trends.
[0118] In the data relationship matching submodule, the network structure is adjusted to match the relationship patterns between data based on state transition analysis. The core of this submodule is to use Bayesian parameter estimation to optimize the network structure. Bayesian parameter estimation is a statistical method that uses Bayesian theorem to update the estimation of parameters that determine the weights of nodes (representing different behaviors or states) and edges (representing the relationship between behaviors or states) in the network. In practice, the initial weights of each node and edge in the network are first adjusted according to the results of the hidden Markov model. Then, by analyzing the relationship patterns in the data, these weights are continuously updated using Bayesian methods to more accurately reflect the relationships in the data. For example, if the data shows that the correlation between two user behaviors increases, the weights of the edges between the nodes representing these two behaviors in the network will increase accordingly. Through these steps, a data relationship matching analysis is generated, which provides an important basis for understanding and utilizing complex relationship patterns in data.
[0119] In the model update submodule, the gradient boosting tree algorithm is applied to adjust the model based on data relationship matching analysis and state transition analysis. Gradient boosting tree is a powerful machine learning algorithm that can optimize the prediction model by building and combining multiple decision trees. In this submodule, the learning rate is set to 0.05 and the maximum depth is set to 4. These parameters are used to control the speed and complexity of model training. The learning rate determines the contribution of each tree to the final model, while the maximum depth limits the growth of the tree to prevent overfitting. Through parameter adjustment, the algorithm can effectively reflect the current data flow and behavior trends of the model. For example, if a user behavior pattern is found to be strongly correlated with a high-risk state, the model will increase the weight of the tree related to this behavior. Finally, an adaptive authentication model is generated, which can dynamically adapt to changes in data flow, including providing powerful tools for identifying and responding to potential risks.
[0120] In the authentication analysis system based on big data, the credit risk index and user behavior provide insights into the user's login frequency, operation type, credit record, etc. For example, one user frequently logs in but performs few operations in a short period of time, while another user has a bad credit record. In the network dynamic adjustment submodule, these data are used to build a hidden Markov model to analyze the state transition of user behavior. In the data relationship matching submodule, the network structure is optimized through Bayesian parameter estimation, highlighting the relationship between frequent logins and bad credit records, for example. In the model update submodule, the gradient boosting tree algorithm is used to adjust the model to reflect trends such as high risk for users who log in frequently. This adaptive authentication model can provide important support for credit risk management and user behavior analysis.
[0121] See also Figure 8,The certification result verification module includes a learning strategy synthesis submodule, a result verification submodule, and a comprehensive score generation submodule;
[0122] The learning strategy synthesis submodule is based on the adaptive authentication model and uses the random forest algorithm to analyze and classify data features. The number of trees in the random forest algorithm is set to 100, the maximum depth is 10, and the Gini impurity is used as the splitting criterion to evaluate feature importance and make classification decisions. It analyzes data features and identifies key classification criteria to generate feature analysis results.
[0123] The result verification submodule uses a gradient boosting machine to perform deep learning and analysis on data features based on the feature analysis results. The gradient boosting machine sets the learning rate to 0.1, the maximum depth to 5, and the loss function to deviance. It refines data features and identifies deep data patterns through iterative optimization, thereby generating deep feature analysis results.
[0124] The comprehensive score generation submodule is based on the results of deep feature analysis, adopts an ensemble learning method, combines multiple differentiation models, integrates the random forest, support vector machine and gradient boosting decision tree algorithms of the Scikit-Learn library, sets the weights and parameters of multiple models, uses the cross_val_score function for cross-validation, and uses a soft voting mechanism to combine the probability estimates of multiple model outputs to calculate the average probability score and generate a comprehensive certification score.
[0125] In the learning strategy synthesis submodule, the data features are carefully analyzed and classified using the random forest algorithm through the adaptive certification model. Random forest is an ensemble learning method that improves the accuracy and robustness of the model by building multiple decision trees. In the implementation of this submodule, the data is first formatted into a structured dataset, where each row represents a sample and each column represents a feature. The number of trees is set to 100 when the random forest algorithm is initialized, which means that 100 decision trees are built, and each tree is trained independently using a different random subset of the sample dataset. The maximum depth is set to 10 to limit the growth of the tree and prevent overfitting. During the tree construction process, Gini impurity is used as the splitting criterion, which is a method to measure the quality of the dataset split and aims to reduce uncertainty. Each split is to divide the dataset into purer subsets. At each node of the decision tree, the algorithm selects the best features and split points to maximize the purity gain. As the tree grows, the algorithm has a deeper understanding of the dataset and gradually identifies key classification criteria. Feature importance assessment is a key step in random forests, which determines the importance of each feature by analyzing its role in the decision tree. After the evaluation, the algorithm is able to identify the features that have the greatest impact on the model's predictive performance. This process not only enhances the interpretability of the model, but also provides important information for subsequent data processing and feature selection. Classification decisions are made by aggregating the predictions of all trees. Each tree gives a prediction, and the final classification decision is based on majority voting or average probability. This approach reduces the error generated by a single tree and improves the accuracy and reliability of the overall model. The final output of this submodule is a data model that has been processed with feature analysis and classification decisions. This model can reveal key features and patterns in the data, providing a solid foundation for subsequent data analysis and decision-making.
[0126] In the result verification submodule, based on the feature analysis results of the previous submodule, the gradient boosting machine (GBM) is used to deeply learn and analyze the data features. The gradient boosting machine is a powerful machine learning technique that creates a strong learner by building and combining multiple weak learners (decision trees). In this process, each new tree model is trained on the errors of the previous tree model, and the model gradually improves its performance through continuous iterative optimization. The learning rate is set to 0.1 to control the impact of each tree on the final model. A smaller learning rate means that the model requires more trees to train, but can get better performance. The maximum depth is set to 5, which helps control the complexity of the model and prevent overfitting. The loss function selects deviance, which is a loss function used for classification problems that aims to quantify the deviation between the predicted value and the actual value. In each iteration, the gradient boosting machine models the residuals of the current model and adds new trees to correct these residuals. This process is repeated until the predetermined number of iterations is reached or the model performance is no longer significantly improved. In this way, the gradient boosting machine can gradually refine and strengthen the data features and identify deep data patterns. The output of this submodule is a deep feature analysis model that can identify and interpret data features with higher accuracy. The results of this deep learning are crucial for understanding complex data structures and relationships, providing deep insights for subsequent data applications and decision-making.
[0127] In the comprehensive score generation submodule, based on the results of deep feature analysis, an ensemble learning method is used to combine multiple differentiated models. Ensemble learning is a method that improves the overall prediction performance by combining multiple models. In this submodule, an integration of the random forest, support vector machine (SVM) and gradient boosted decision tree (GBDT) algorithms of the Scikit-Learn library is used. These models are trained independently, and each model has a different understanding and interpretation of the data. By combining these models, more comprehensive and accurate results can be obtained. The weight and parameter setting of multiple models are the key parts of ensemble learning. The weight determines the influence of each model in the final prediction. The parameter setting needs to be adjusted according to the characteristics of each model and the characteristics of the data to achieve the best performance. Cross-validation using the cross_val_score function is to evaluate the performance and stability of the model. Cross-validation provides a comprehensive evaluation of model performance by dividing the dataset into multiple parts and then training and testing the model on different subsets. The soft voting mechanism is a technique in ensemble learning that combines the probability estimates of multiple model outputs instead of simply making a final classification decision. This approach allows the model to leverage the knowledge of all models rather than relying solely on the judgment of a single model. The output of the comprehensive score generation submodule is a certification comprehensive score that reflects the results of comprehensive consideration of multiple models. This score is not only based on in-depth analysis of data features, but also incorporates the unique perspectives and advantages of different models, providing a solid basis for the final certification decision.
[0128] In a student learning behavior certification analysis system for an online education platform, three submodules work together to deeply analyze students' learning behavior. First, the learning strategy synthesis submodule uses the random forest algorithm to analyze structured student data. These data include learning time (hours), course completion rate (percentage), test scores (scores), discussion area activity (number of times), and video viewing habits (percentage). By setting 100 trees and a maximum depth of 10, the algorithm evaluates the impact of each feature on student learning outcomes and generates a feature importance report to reveal which behavior patterns are related to learning outcomes. Next, the result verification submodule uses a gradient boosting machine to deeply analyze these features. Setting the parameter configuration of the learning rate to 0.1 and the maximum depth to 5, coupled with the application of the deviance loss function, enables the module to more accurately identify the key factors that affect students' learning outcomes, thereby generating a deep feature analysis model. Finally, the comprehensive score generation submodule uses an ensemble learning method to combine the outputs of the random forest, support vector machine, and gradient boosting decision tree algorithms to comprehensively consider the predictions of each model. Through cross-validation and soft voting mechanisms, this module generates a comprehensive score of student learning behavior based on multi-dimensional analysis. This score not only reflects students’ performance in various aspects, but also provides valuable insights for educational platforms to help improve teaching methods and provide more personalized learning recommendations for students.
[0129] See also Fig. 9 ,The chain data integrity module includes a blockchain construction submodule, a data immutability assurance submodule, and an integrity proof generation submodule;
[0130] The blockchain construction submodule is based on the comprehensive certification score, uses blockchain technology, and creates a blockchain structure through programming language. Each block uses the SHA-256 hash algorithm to generate a unique hash value, processes the data record, the hash value of the previous block, and the current data into a string, and applies the hash algorithm to convert it into a unique hash value of a fixed length, and continuously connects each block in series to generate an encrypted data chain;
[0131] The data immutability guarantee submodule is based on the encrypted data chain. It uses a combination of hash functions and timestamps through programming to lock the content of each block, apply hash functions to the data and timestamps in each block to generate the current hash value, assist in data recording in the blockchain network, and generate data immutability records;
[0132] The integrity proof generation submodule is based on data invariance records and adopts the blockchain integrity verification program to check the data and structure of each block in the blockchain, and uses programming logic to verify the integrity and change item status of each block in the chain to generate a chain integrity proof.
[0133] In the blockchain building submodule, a blockchain structure is implemented through programming language to store and protect the certification comprehensive score data. The process starts with defining the blockchain structure, where each block contains multiple data items, such as score data, timestamp, and hash value of the previous block. The data items are stored in string format for subsequent processing. Next, the SHA-256 hash algorithm, a widely recognized cryptographic hash function, is used to generate a unique and fixed-length hash value for each block. The generation of the hash value involves combining the data record in the block, the hash value of the previous block, and the current timestamp into a string, which is then converted into a hash value through the SHA-256 algorithm. This hash value is unique, and even a slight change in the data will result in a completely different hash value generation, thereby ensuring the security and immutability of the data. In the process of continuously connecting each block in series, the creation of each new block depends on the hash value of the previous block, forming an irreversible chain structure. In this way, once the data is recorded in a block, it cannot be changed or deleted, ensuring the immutability and security of the data.
[0134] In the data immutability guarantee submodule, the combination of hash functions and timestamps is implemented through programming to further enhance the security of the blockchain. In this process, the hash function is applied to the data and timestamp in each block to generate the hash value of the current block. This operation includes not only the data content in the block, but also the hash value of the previous block and the timestamp of the current block, ensuring that every part of the blockchain is sealed and protected. The generated hash value is used as a unique identifier for each block, ensuring that every piece of data on the chain is original and has not been tampered with. This mechanism ensures that any attempt to change the data of an existing block in the chain will result in a mismatch in the hash value, which can be quickly detected. In this way, the data records in the blockchain network are solidified, generating immutable data records.
[0135] In the integrity proof generation submodule, a blockchain integrity verification program is used. The task of this program is to perform an exhaustive check of the data and structure of each block in the blockchain. Through programming logic, the integrity and change item status of each block in the chain are verified. This includes checking whether the hash value of each block is correct, that is, whether it matches the hash value of the block content and the previous block. At the same time, the program also checks whether the links between blocks are complete and ensure that no blocks have been removed or tampered with. Any inconsistencies in this process will be identified, thereby ensuring the data integrity and reliability of the entire chain. Finally, this submodule generates a chain integrity proof, which is the final confirmation of the integrity and security of the entire blockchain and its contents.
[0136] Assume a big data-based authentication analysis system that processes data including student grades, attendance records, online test scores, etc. These data items exist in the form of numbers and text, such as student ID, grades (scores), attendance (number of days of attendance), and online test scores (scores). In the blockchain construction submodule, each student's data is encoded into a block, which contains the student's ID, grades, attendance records, and online test scores. Through the SHA-256 hash algorithm, each block generates a unique hash value. In the data immutability assurance submodule, a hash function is applied to each block to generate a new hash value and combine it with the timestamp to ensure that the data of each block cannot be changed. Finally, in the integrity proof generation submodule, the blockchain integrity verification program checks the data and structure of each block, ensures the integrity and reliability of the entire chain, and generates a chain integrity proof, which provides a safe and reliable way for educational institutions to manage and authenticate student data.
[0137] See also Fig.10 , a big data-based authentication analysis method, the big data-based authentication analysis method is performed based on the above-mentioned big data-based authentication analysis system, and includes the following steps:
[0138] S1: Based on the original authentication dataset, a dynamic time warping algorithm is used to perform shape matching and alignment of time series, and a sequence clustering method is used to classify and extract patterns from time series data, analyze the temporal regularity of user behavior, extract key temporal features, and generate a temporal behavior feature set;
[0139] S2: Based on the temporal behavior feature set, the isolation forest algorithm is used to score each instance for anomaly. By analyzing and identifying abnormal behaviors that differ from regular user behavior patterns, a security anomaly assessment is generated.
[0140] S3: Based on security anomaly assessment, neural networks are used to analyze multidimensional risk factors, and support vector machines are used to classify and score risk levels, conduct a comprehensive quantitative assessment of credit risk, and generate a credit risk index;
[0141] S4: Based on the original authentication data, the data preprocessing technology is used to organize and standardize the behavior data, and the prefix span algorithm is applied to deeply mine and analyze the user behavior patterns to generate user behavior insight results;
[0142] S5: Based on the credit risk index and user behavior insight results, the hidden Markov model is used to model the state transition of the data stream, and the dynamic Bayesian network is updated according to the model results to match the current data pattern, thereby generating an adaptive authentication model;
[0143] S6: Based on the adaptive authentication model, random forest, gradient boosting machine and ensemble learning methods are used to analyze and verify data features, and then blockchain technology is used to encrypt and process authentication data to generate chain integrity proof.
[0144] By accurately analyzing the temporal regularity of user behavior and extracting key time features, the system has made significant progress in understanding user behavior patterns, which not only improves the accuracy of analysis, but also lays the foundation for subsequent abnormal behavior identification and risk assessment. In particular, the introduction of the isolation forest algorithm has effectively improved the ability to identify abnormal behavior and provided a powerful tool for preventing fraud and security threats. The comprehensive credit risk assessment combined with neural networks and support vector machines makes risk management more accurate and comprehensive, which helps to better control and manage credit risks. The application of the prefix span algorithm further deepens the insight into user behavior and provides data support for providing personalized services and improving user satisfaction. Through the establishment of an adaptive authentication model, the system can flexibly respond to changes in behavior patterns and enhance security and reliability. Finally, by combining random forests, gradient boosting machines and blockchain technology, the system has not only improved the accuracy and efficiency of data analysis, but also made significant contributions to data security and invariance assurance, enhancing the trust and reliability of the entire system.
[0145] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. A big data-based authentication analysis system, characterized by: The system includes a behavior time analysis module, a security anomaly identification module, a credit risk quantification module, a user behavior mining module, an authentication data matching module, an authentication result verification module, and a chain data integrity module; The behavior time analysis module uses a dynamic time warping algorithm to match and align the shape of the time series based on the original authentication data set, and applies a sequence clustering method to classify and extract patterns from the time series data, analyze the temporal regularity of user behavior, and generate a temporal behavior feature set; The behavior time analysis module includes a time dependency analysis submodule, a sequence pattern recognition submodule, and a behavior feature extraction submodule; The time dependency analysis submodule uses a dynamic time warping algorithm to perform shape matching and alignment of time series based on the original authentication data set, constructs a distance matrix by calculating the distance between each pair of time points, and uses dynamic programming technology to capture the path with the minimum cumulative distance, performs optimal matching between sequences, and generates a time series alignment set; The sequence pattern recognition submodule classifies and extracts patterns from time series data using a sequence clustering method based on a time series alignment set, selects K-means or hierarchical clustering, selects the number of clusters and a distance metric, classifies time series data, identifies patterns and rules, and generates a sequence pattern feature set; The behavior feature extraction submodule uses the principal component analysis method to reduce the dimension of the extracted features based on the sequence pattern feature set, extracts the principal components by calculating the covariance matrix of the data, and then reduces the dimension of the data according to the principal components, extracts the key behavior features, and generates a time behavior feature set; The security anomaly identification module uses the isolation forest algorithm to perform anomaly scoring on each instance in the feature set based on the time behavior feature set, identifies abnormal behaviors that differ from the normal pattern, and generates security anomaly indicators; The safety anomaly identification module includes a behavior safety analysis submodule, an abnormal pattern identification submodule, and a safety indicator generation submodule; The behavior security analysis submodule uses the isolation forest algorithm to score and identify abnormal behaviors based on the time behavior feature set. An isolation tree is constructed by randomly selecting features and selecting the cut value of the features. The number of constructed trees is set to 100, and the number of randomly selected samples for each tree is set to 256. The average path length from the sample point to the leaf is calculated, and samples with shorter path lengths are considered abnormal, thereby generating an abnormal behavior identification set; The abnormal pattern recognition submodule uses the K-means clustering algorithm to recognize and classify abnormal patterns based on the abnormal behavior recognition set, sets the number of cluster centers to 5, uses the Euclidean distance as the similarity metric, and optimizes the cluster center positions iteratively to group abnormal behaviors into differentiated categories to generate a patterned abnormal analysis set; The safety index generation submodule is based on the patterned anomaly analysis set, uses a scoring system to classify each anomaly pattern into a safety level, sets the safety level range from 1 to 5, where 1 represents low risk, and then generates a safety anomaly index; The credit risk quantification module uses neural networks to analyze multidimensional risk factors based on security anomaly indicators, classifies and scores risk levels through support vector machines, and conducts a comprehensive quantitative assessment of credit risks to generate a credit risk index; The credit risk quantification module includes a risk measurement submodule, a risk dynamic prediction submodule, and an index output submodule; The risk measurement submodule uses a neural network to analyze multidimensional risk factors based on security anomaly indicators. The constructed neural network includes two hidden layers, each with 128 neurons. The ReLU activation function is used to adjust the network weights through the gradient descent algorithm. The learning rate is set to 0.01 and the batch size is 32. The nonlinear relationship and intrinsic pattern between risk factors are captured to generate a risk factor analysis set. The risk dynamic prediction submodule uses a support vector machine to classify and score risk levels based on a risk factor analysis set. The SVM model selects a radial basis function as a kernel function, the C parameter is set to 1.0 to balance training error and model complexity, and the gamma parameter is set to 0.1 to determine the influence range of the kernel function. The risk factors are classified and processed to generate a risk prediction analysis set. The index output submodule is based on the risk prediction analysis set, uses the weighted average method to comprehensively refer to multiple types of risk factors, sets weights for each risk factor, performs weighted calculations based on the degree of impact and importance on credit risk, and then generates a credit risk index; The user behavior mining module uses data preprocessing technology to organize and standardize behavior data based on the original authentication data, and applies the prefix span algorithm to deeply mine and analyze user behavior patterns to generate user behavior insights; The user behavior mining module includes a behavior pattern mining submodule, a user habit analysis submodule, and a behavior insight building submodule; The behavior pattern mining submodule uses Python's Pandas library to clean the data based on the original authentication data, and applies the Apriori algorithm to mine association rules, setting the minimum support to 0.01 and the minimum confidence to 0.
5. It also uses the mlxtend library to mine frequent itemsets and association rules, analyze user behavior patterns, and generate in-depth analysis of behavior patterns. The user habit analysis submodule uses a machine learning algorithm based on in-depth analysis of behavioral patterns, uses the Scikit-learn library to perform K-means cluster analysis, sets the number of clusters to 5, uses the fit_predict method to cluster user habit data, analyzes the behavioral habits of multiple user groups, and uses principal component analysis to reduce data dimensionality, retains the principal components, and generates user habit analysis; The behavior insight building submodule is based on comprehensive analysis of user habits, uses NumPy and Pandas libraries to merge and reorganize data, combines behavior patterns and user habit data, and uses the support vector machine algorithm to classify behavior. It uses the SVC function in the Scikit-learn library, sets the kernel function to the radial basis function, and the C parameter to 1.
0. The fit method is used to train the model, adjust and optimize the classification results, and generate user behavior insights; The authentication data matching module uses a hidden Markov model to model the state transition of the data stream based on the credit risk index and user behavior insights, and updates the dynamic Bayesian network according to the model results to match the current data pattern and generate an adaptive authentication model; The authentication data matching module includes a network dynamic adjustment submodule, a data relationship matching submodule, and a model updating submodule; The network dynamic adjustment submodule uses a hidden Markov model to model the state transition of data streams based on the credit risk index and user behavior insights. The number of states in the model is set to 5, and the transition probability is dynamically adjusted according to the behavior frequency in the data stream, and the state change trend in the data stream is analyzed to generate a state transition analysis. The data relationship matching submodule is based on state transition analysis, adjusts the network structure according to the results of the hidden Markov model, resets the weights of nodes and edges, matches the relationship pattern between data, optimizes the network structure using Bayesian parameter estimation, and generates data relationship matching analysis; The model update submodule adjusts the model based on data relationship matching analysis and state transition analysis, applies the gradient boosting tree algorithm, sets the learning rate to 0.05 and the maximum depth to 4, and generates an adaptive authentication model by adjusting the parameters to reflect the current data flow and behavior trend of the model; The certification result verification module is based on an adaptive certification model, applies a random forest algorithm to analyze and classify data features, and uses a gradient boosting machine to perform deep learning and analysis of data features, and then cross-validates the certification results through an integrated learning method to generate a comprehensive certification score; The authentication result verification module includes a learning strategy synthesis submodule, a result verification submodule, and a comprehensive score generation submodule; The learning strategy synthesis submodule is based on the adaptive authentication model and uses the random forest algorithm to analyze and classify data features. The number of trees in the random forest algorithm is set to 100, the maximum depth is 10, and the Gini impurity is used as the splitting criterion to perform feature importance evaluation and classification decisions, analyze data features and identify key classification criteria, and generate feature analysis results; The result verification submodule uses a gradient boosting machine to perform deep learning and analysis on data features based on the feature analysis results. The gradient boosting machine sets the learning rate to 0.1, the maximum depth to 5, and the loss function to deviance. It refines data features and identifies deep data patterns through iterative optimization, thereby generating deep feature analysis results. The comprehensive score generation submodule adopts an ensemble learning method based on the results of deep feature analysis, combines multiple differentiation models, integrates the random forest, support vector machine and gradient boosting decision tree algorithms of the Scikit-Learn library, sets the weights and parameters of multiple models, uses the cross_val_score function for cross-validation, and uses a soft voting mechanism to combine the probability estimates of multiple model outputs to calculate the average probability score and generate a comprehensive certification score; The chain data integrity module uses blockchain technology to encrypt the authentication data based on the comprehensive authentication score, and stores the encrypted data in the blockchain network for continuous chain recording, thereby generating a chain integrity certificate; The chain data integrity module includes a blockchain construction submodule, a data invariance assurance submodule, and an integrity proof generation submodule; The blockchain construction submodule is based on the comprehensive certification score, uses blockchain technology, and creates a blockchain structure through a programming language. Each block uses the SHA-256 hash algorithm to generate a unique hash value, processes the data record, the hash value of the previous block, and the current data into a string, applies the hash algorithm to convert it into a unique hash value of a fixed length, and continuously concatenates each block to generate an encrypted data chain; The data invariance guarantee submodule is based on the encrypted data chain, and uses a combination of hash functions and timestamps through programming to lock the content of each block, apply a hash function to the data and timestamp in each block to generate the current hash value, assist in data recording in the blockchain network, and generate data invariance records; The integrity proof generation submodule uses a blockchain integrity verification program based on data invariance records to check the data and structure of each block in the blockchain, uses programming logic to verify the integrity and change item status of each block in the chain, and generates a chain integrity proof.
2. The authentication analysis system based on big data according to claim 1, characterized in that: The time behavior feature set includes the time pattern of user behavior, key behavior time points, and statistical characteristics of time series; the security anomaly indicators include abnormal behavior scores, behavior deviations, and frequencies of abnormal behaviors; the credit risk index includes risk levels, key risk factors, and risk trend forecasts; the user behavior insights include user behavior patterns, key behavior characteristics, and behavior change trends; the adaptive authentication model includes network structure adjustment parameters, data relationship weights, and model adaptability indicators; the authentication comprehensive score includes data consistency scores, authentication accuracy, and verification efficiency indicators; the chain integrity proof includes blockchain record information, data integrity check codes, and data consistency proofs.
3. A big data-based authentication analysis method, characterized in that: The big data-based authentication analysis system according to any one of claims 1 to 2 comprises the following steps: Based on the original authentication data set, a dynamic time warping algorithm is used to perform shape matching and alignment of time series, and a sequence clustering method is used to classify and extract patterns from time series data, analyze the temporal regularity of user behavior, extract key temporal features, and generate a temporal behavior feature set; Based on the temporal behavior feature set, an isolation forest algorithm is used to perform anomaly scoring on each instance, and a security anomaly assessment is generated by analyzing and identifying abnormal behaviors that differ from regular user behavior patterns; Based on the security anomaly assessment, a neural network is used to analyze multidimensional risk factors, and risk levels are classified and scored through a support vector machine to conduct a comprehensive quantitative assessment of credit risk and generate a credit risk index; Based on the original authentication data, the data preprocessing technology is used to organize and standardize the behavior data, and the prefix span algorithm is applied to deeply mine and analyze the user behavior patterns to generate user behavior insight results; Based on the credit risk index and user behavior insight results, a hidden Markov model is used to model the state transition of the data stream, and a dynamic Bayesian network is updated according to the model results to match the current data pattern, thereby generating an adaptive authentication model; Based on the adaptive authentication model, random forest, gradient boosting machine and ensemble learning methods are used to analyze and verify data features, and then blockchain technology is used to encrypt and process authentication data to generate a chain integrity certificate.
