Dichotomy method and system for human pairwise comparison in simulation decision-making process
The binary classification model is constructed through data preprocessing, random forest feature extraction and multiplication priority simulation decision-making methods, which solves the problems of inaccurate feature extraction, high computational complexity and noise sensitivity in traditional methods, and achieves efficient and stable data classification.
Patent Information
- Application Number
- CN202510498454.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-08-08
AI Technical Summary
The traditional binary classification method is not accurate enough to extract the feature of data, resulting in low classification accuracy, high computational complexity, difficulty in quickly processing large-scale data, and sensitive to noise and outliers, affecting the stability and reliability of classification.
A binary classification model is constructed using data preprocessing, random forest feature extraction and multiplication priority simulation decision-making methods, combined with cross-validation technology to optimize model parameters, through data cleaning, feature normalization and standardization processing, random forests are used to evaluate feature importance, and multiplication priority preference classifier is constructed to optimize model performance.
It significantly improves the accuracy and reliability of classification, can quickly process large-scale data, enhances the stability and applicability of the method, and is suitable for many fields such as medical care, finance, and image recognition.
Smart Images
Figure CN120448988A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of data classification, and in particular relates to a binary classification method and system for simulating human pairwise comparison in a decision-making process. Background Art
[0002] With the rapid development of information technology, data classification plays a vital role in numerous fields. Accurately classifying data into two categories is a challenging and practical task in many practical applications. For example, in medical diagnosis, it is necessary to distinguish whether a patient's condition is benign or malignant; in finance, it is necessary to determine whether a transaction is risky; and in image recognition, it is necessary to determine whether an object in an image belongs to one of two specific categories.
[0003] Traditional binary classification methods often have limitations. Some methods may not accurately extract data features, resulting in low classification accuracy; others may have high computational complexity, making it difficult to quickly process large amounts of data; and still others may be sensitive to noise and outliers, affecting the stability and reliability of classification.
[0004] Through the above analysis, the problems and defects of the existing technology are as follows: the traditional binary classification method is not accurate enough in extracting data features, resulting in low classification accuracy; the computational complexity is high, making it difficult to quickly process large-scale data; it is sensitive to noise and outliers, affecting the stability and reliability of classification. Summary of the Invention
[0005] To overcome the problems existing in the related art, the embodiments disclosed in the present invention provide a binary classification method and system that simulates human pairwise comparison in the decision-making process. The technical solution is as follows:
[0006] The present invention is implemented by simulating a binary classification method of human pairwise comparison in a decision-making process, comprising the following steps:
[0007] S1, data preprocessing: clean the input data to remove noise and outliers; perform feature normalization and standardization to make different features have the same scale;
[0008] S2, feature extraction: random forest method is introduced to extract data features;
[0009] S3, classification model construction: using multiplication priority to simulate the human pairwise comparison method in the decision-making process to build a binary classification model;
[0010] S4, model training and optimization: Use labeled training data to train the classification model, use cross-validation techniques to evaluate the performance and generalization ability of the model, and select the model parameters with the best performance;
[0011] S5, classification result output: input the data to be classified into the trained model to obtain the binary classification result.
[0012] In step S1, feature normalization and standardization processing are performed, including: using mean normalization or standard deviation normalization method to make different features have the same scale, the expression is:
[0013]
[0014] Where X is the array of sample data sets, max(X) is the maximum value of X by column, and min(X) is the minimum value of X by column.
[0015] In step S2, the random forest is composed of multiple decision trees. During the training process of the constructed random forest model, the importance of different features is evaluated by training multiple decision trees, thereby mining the key feature information in the data; the Random Forest Classifier Random Forest Classifier in the sklearn library is imported, and the random forest classifier is created and parameters are set, and then training is performed.
[0016] Random forest returns an indicator value of the importance of each feature, retains features greater than a given threshold and uses them as the final feature set, and discards features less than a given threshold.
[0017] In step S3, a binary classification model is constructed, including:
[0018] Assume that m samples are X={x1,x2…x m}, each sample has n features, and the i-th sample is The multiplication priority between the first and second samples at the jth feature is:
[0019]
[0020] Where >>> is the defined multiplication priority, h -1 is the inverse function of h, is the jth feature of sample x1, is the j-th feature of sample x2;
[0021] Through nonlinear weighted aggregation operator Aggregate the priorities between the individual features of the two samples to obtain the comprehensive multiplication priority between the two samples, which is expressed as:
[0022]
[0023] Where, ω j is the weight coefficient to be determined; a j for
[0024] Using the comprehensive multiplication priority of pairwise comparisons between any two samples, a fuzzy preference relationship with multiplication consistency is constructed; the ranking index of each sample is obtained, and the expression is:
[0025]
[0026] Where, ω is the weight parameter to be determined, m is the number of samples, and n is the number of features.
[0027] Introducing bias Assist classification decision-making, determine the classification rules according to the relationship between the ranking index and the bias item: if the ranking index takes precedence over the bias item, that is, Classify the email as positive; otherwise Classified as negative;
[0028] With the help of nonlinear aggregation operator And the sample label value, the two classification rules are aggregated into one rule to form a multiplication priority preference classifier, the expression is:
[0029]
[0030] Where, L t is the degree to which the tth sample belongs to the positive or negative class, y (t) is the t-th sample label, is the bias term.
[0031] In step S3, the model is optimized by adjusting the parameters of the model to optimize the performance of the model. The parameters of the model include weight ω and bias term
[0032] In step S4, model training and optimization include: using an optimization algorithm to minimize the classification error; the optimization algorithm uses a stochastic gradient descent algorithm, and the stochastic gradient descent algorithm includes:
[0033] Determine the loss function: Use nonlinear logical operations to interpret the aggregated rules and derive the loss function as follows:
[0034]
[0035] The initial values of the learning rate and the number of training rounds are set to 0.1 and 100 respectively. For each training sample x t Calculate the loss function J t Regarding the gradient of the model parameters, the formula is:
[0036]
[0037] Where ω is the weight,
[0038] Update the model parameters ω and bias terms by calculating the gradient After reaching the scheduled training rounds, the final evaluation results of the model are calculated;
[0039] The data set is randomly shuffled and divided into 5 subsets. The ith subset is selected as the validation set, and the other 4 subsets are selected as the training set. The training set is used to train model 1 to obtain model M. i , use the validation set to evaluate the trained model and calculate the evaluation indicators; after 5 rounds of training and validation, 5 evaluation indicators are obtained; by observing the changes of these indicators on the validation set, it is determined whether the model is overfitting or underfitting.
[0040] In step S5, after obtaining the binary classification result using the trained model, the category label or the probability value belonging to each category is output.
[0041] Another object of the present invention is to provide a binary classification system for simulating human pairwise comparison in a decision-making process, the system being used to regulate the binary classification method for human pairwise comparison in the decision-making process, the system comprising:
[0042] The data preprocessing module is used to clean the input data, remove noise and outliers, and perform feature normalization and standardization to make different features have the same scale;
[0043] Feature extraction module, used to introduce random forest method to extract data features;
[0044] The classification model building module is used to build a binary classification model by simulating the human pairwise comparison method in the decision-making process using multiplication priority;
[0045] The model training and optimization module is used to train the classification model using labeled training data, evaluate the performance and generalization ability of the model using cross-validation techniques, and select the model parameters with the best performance;
[0046] The classification result output module is used to input the data to be classified into the trained model to obtain the binary classification results.
[0047] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:
[0048] 1. Improved Classification Accuracy: This method combines traditional feature engineering methods with the random forest method to extract data features more comprehensively and accurately. Extensive experiments and practical applications have demonstrated that this method significantly improves classification accuracy and reliability compared to single methods or other conventional combinations.
[0049] 2. Rapidly process large-scale data: Optimized classification models and algorithms can quickly process large-scale data while ensuring accuracy, thereby improving work efficiency.
[0050] 3. Enhanced stability and reliability: The application of data preprocessing and regularization techniques makes the method more robust to noise and outliers, and improves the stability and reliability of classification.
[0051] 4. Wide applicability: This binary classification method can be applied to multiple fields, such as medicine, finance, image recognition, etc., and has wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;
[0053] Figure 1 This is a flow chart of a binary classification method based on multiplication priority simulation pairwise comparison provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0054] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0055] The innovation of the present invention lies in: simulating the decision maker's preference decision-making induction binary classification method based on multiplication priority, simulating the decision-making behavior of humans when doing binary classification, which is not mentioned in existing methods.
[0056] Example 1, spam problem: Figure 1 As shown, the binary classification method based on multiplication priority simulation pairwise comparison provided by the embodiment of the present invention includes the following steps:
[0057] S1, data preprocessing: clean the input data to remove noise and outliers; perform feature normalization and standardization to make different features have the same scale;
[0058] Use statistical methods or machine learning-based anomaly detection techniques to identify and remove anomalous data points. Make different features have the same scale to facilitate subsequent classification calculations. For example, use mean normalization or standard deviation normalization.
[0059] Perform feature normalization to make different features have the same scale, so as to better perform subsequent classification calculations. For example, use mean normalization or standard deviation normalization to make different features have the same scale. The expression is:
[0060]
[0061] Where X is the array of sample data sets, max(X) is the maximum value of X by column, and min(X) is the minimum value of X by column.
[0062] S2, feature extraction: random forest method is introduced to extract data features;
[0063] The random forest method is introduced for feature extraction. Random forest is an ensemble learning method composed of multiple decision trees. By training multiple decision trees in a random forest, the importance of different features can be assessed. Highly important features are selected as the final feature set, further improving feature quality and classification accuracy. The random forest returns an importance index for each feature. Features with an importance greater than a given threshold are retained as the final feature set, while features with an importance less than a given threshold are discarded.
[0064] The random forest is composed of multiple decision trees. During the training process of the random forest model constructed by the present invention, the importance of different features can be evaluated by training multiple decision trees, thereby mining key feature information in the data.
[0065] Import the random forest classifier RandomForestClassifier from the sklearn library, create a random forest classifier, set parameters, and then train it:
[0066] rf=RandomForestClassifier(n_estimators=100,random_state=42)
[0067] rf.fit(X,y)
[0068] Here we create a random forest classifier with 100 trees, set the random seed, and train it using the given data. We calculate the feature importance:
[0069] importances=rf.feature_importances_
[0070] This will output the importance score of each feature. We can then extract features based on the importance score. For example, here we set a threshold of 0.01, use the np.where function to find the indices of features whose importance scores are greater than the threshold, and then extract the corresponding features from the dataset based on these indices. (In practice, the threshold is a hyperparameter, obtained through hyperparameter tuning). Here we assume threshold = 0.01
[0071] selected_features_indices=np.where(importances>threshold)[0]
[0072] X_selected=X[:,selected_features_indices]
[0073] S3, classification model construction: using multiplication priority to simulate the human pairwise comparison method in the decision-making process to build a binary classification model;
[0074] The Python library Optuna is used to adjust the model's hyperparameters, such as the number of trees in the random forest, to optimize the model's performance.
[0075] Build a binary classification model and perform model optimization, including:
[0076] Assume that m samples are X={x1,x2…x m}, each sample has n features, and the i-th sample is The multiplication priority between the first and second samples at the jth feature is:
[0077]
[0078] Where >>> is the defined multiplication priority, h -1 is the inverse function of h, is the jth feature of sample x1, is the j-th feature of sample x2;
[0079] Through nonlinear weighted aggregation operator Aggregate the priorities between the individual features of the two samples to obtain the comprehensive multiplication priority between the two samples, which is expressed as:
[0080]
[0081] Where, ω j is the weight coefficient to be determined; a j for
[0082] As shown in Table 1 below, the comprehensive multiplicative priority of pairwise comparisons between any two samples is used to construct a fuzzy preference relationship with multiplicative consistency;
[0083] Table 1 Fuzzy preference relations of multiplicative consistency
[0084]
[0085] Find the ranking index of each sample, the expression is:
[0086]
[0087] Where, ω is the weight parameter to be determined, m is the number of samples, and n is the number of features.
[0088] Introducing bias Assist classification decision-making, determine the classification rules according to the relationship between the ranking index and the bias item: if the ranking index takes precedence over the bias item, that is, Classify the email as positive; otherwise Classified as negative;
[0089] With the help of nonlinear aggregation operator And the sample label value, the two classification rules are aggregated into one rule to form a multiplication priority preference classifier, the expression is:
[0090]
[0091] Where, L t is the degree to which the tth sample belongs to the positive or negative class, y (t) is the t-th sample label, is the bias term.
[0092] S4, model training and optimization: Use labeled training data to train the classification model, use cross-validation techniques to evaluate the performance and generalization ability of the model, and select the model parameters with the best performance;
[0093] The classification model is trained using labeled training data. A suitable optimization algorithm, such as stochastic gradient descent (SGD) or Adam optimizer, is used to minimize the classification error.
[0094] Introduce regularization techniques, such as L1 regularization and L2 regularization, to prevent model overfitting.
[0095] Cross-validation and other techniques are used to evaluate the performance of the model and select the model parameters with the best performance.
[0096] Determine the loss function: Use nonlinear logical operations to interpret the aggregated rules and derive the loss function as follows:
[0097]
[0098] The initial values of the learning rate and the number of training rounds are set to 0.1 and 100 respectively. For each training sample x t Calculate the loss function J t Regarding the gradient of the model parameters, the formula is:
[0099]
[0100] Where, ω is the weight;
[0101] Update the model parameters by calculating the gradient, reach the predetermined training round, and calculate the final evaluation result of the model;
[0102] The data set is randomly shuffled and divided into 5 subsets. The i-th subset is selected as the validation set, and the other 4 subsets except the i-th subset are used as training sets. The training set is used to train model 1 to obtain model M. i , use the validation set to evaluate the trained model and calculate the evaluation indicators; after 5 rounds of training and validation, 5 evaluation indicators are obtained; by observing the changes of these indicators on the validation set, it is determined whether the model is overfitting or underfitting.
[0103] S5, classification result output: input the data to be classified into the trained model to obtain the binary classification result;
[0104] Input the data to be classified into the trained model to obtain a binary classification result. You can output the category label or the probability value of belonging to each category.
[0105] After obtaining the binary classification results using the trained model, the category labels or the probability values belonging to each category are output.
[0106] Example 2, medical diagnosis scenario:
[0107] Collect a large amount of patient medical data, including symptoms, test results, and medical history. After preprocessing this data, perform preliminary feature extraction using a deep learning model and traditional feature engineering methods. Then, use a random forest algorithm to assess feature importance and select key features. Use the binary classification method of our invention to build a classification model to distinguish between benign and malignant conditions.
[0108] Doctors can input new patient data into the trained model to quickly obtain diagnostic results and provide reference for treatment decisions.
[0109] Example 3, financial risk assessment scenario:
[0110] Financial transaction data, including transaction amounts, transaction times, and transaction partners, is collected. This method analyzes this data by first preprocessing it and then applying a combination of feature extraction methods, including random forests, to identify key features. This determines whether the transaction presents risks, such as fraud or credit risk.
[0111] Financial institutions can use this method to monitor transactions in real time and promptly identify and prevent risks.
[0112] Example 4, image recognition scenario:
[0113] Image data is first preprocessed, such as cropping, scaling, and normalization. A deep learning model is then used to extract image features, which are then filtered using a random forest algorithm. Finally, a binary classification method is used to determine whether the objects in the image belong to one of two specific categories, such as distinguishing between images of cats and dogs.
[0114] It can be applied to fields such as intelligent security and autonomous driving to improve the accuracy and efficiency of image recognition.
[0115] Example 5: The binary classification system based on multiplication priority simulation pairwise comparison provided by the embodiment of the present invention specifically includes:
[0116] Data preprocessing module, used to clean and standardize the input data;
[0117] Feature extraction module, used to introduce random forest method to extract data features;
[0118] The classification model building module is used to build a binary classification model and perform model optimization by simulating the decision maker's preference decision using multiplication priority;
[0119] The model training and optimization module is used to train the classification model using labeled training data; cross-validation technology is used to evaluate the performance of the model and select the model parameters with the best performance;
[0120] The classification result output module is used to input the data to be classified into the trained model, obtain the binary classification results, and output the category label or the probability value belonging to each category.
[0121] The above description is only a preferred specific implementation method of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.
Claims
1. A binary classification method that simulates human pairwise comparison in the decision-making process, characterized by: The method comprises the following steps: S1, data preprocessing: clean the input data to remove noise and outliers; perform feature normalization and standardization to make different features have the same scale; S2, feature extraction: random forest method is introduced to extract data features; S3, classification model construction: using multiplication priority to simulate the human pairwise comparison method in the decision-making process to build a binary classification model; S4, model training and optimization: Use labeled training data to train the classification model, use cross-validation techniques to evaluate the performance and generalization ability of the model, and select the model parameters with the best performance; S5, classification result output: input the data to be classified into the trained model to obtain the binary classification result.
2. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 1, characterized in that: In step S1, feature normalization and standardization processing are performed, including: using mean normalization or standard deviation normalization method to make different features have the same scale, the expression is: Where X is the array of sample data sets, max(X) is the maximum value of X by column, and min(X) is the minimum value of X by column.
3. According to the binary classification method for human pairwise comparison in the simulated decision-making process of claim 1, in step S2, the random forest is composed of multiple decision trees. During the training process of the constructed random forest model, the importance of different features is evaluated by training the multiple decision trees, thereby mining key feature information in the data; the Random Forest Classifier Random Forest Classifier in the sklearn library is imported, and the random forest classifier is created and parameters are set, and then training is performed.
4. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 3, characterized in that: Random forest returns an indicator value of the importance of each feature, retains features greater than a given threshold and uses them as the final feature set, and discards features less than a given threshold.
5. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 1, characterized in that: In step S3, a binary classification model is constructed, including: Assume that m samples are X={x1,x2…x m }, each sample has n features, and the i-th sample is The multiplication priority between the first and second samples at the jth feature is: Where >>> is the defined multiplication priority, h -1 is the inverse function of h, is the jth feature of sample x1, is the j-th feature of sample x2; Through nonlinear weighted aggregation operator Aggregate the priorities between the individual features of the two samples to obtain the comprehensive multiplication priority between the two samples, which is expressed as: in, Where, ω j is the weight coefficient to be determined, a j for Using the comprehensive multiplication priority of pairwise comparisons between any two samples, a fuzzy preference relationship with multiplication consistency is constructed; the ranking index of each sample is obtained, and the expression is: Where, ω is the weight parameter to be determined, m is the number of samples, and n is the number of features.
6. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 5, characterized in that: Also includes: Introducing bias terms Assist classification decision-making, determine the classification rules according to the relationship between the ranking index and the bias item: if the ranking index takes precedence over the bias item, that is, Classify the email as positive; otherwise p t , classified as negative class; With the help of nonlinear aggregation operator and sample label values, aggregate the two classification rules into one rule to form a multiplication priority preference classifier, which is expressed as: Where, L t is the degree to which the tth sample belongs to the positive or negative class, y (t) is the t-th sample label, is the bias term.
7. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 1, characterized in that: In step S3, the model is optimized by adjusting the parameters of the model to optimize the performance of the model. The parameters of the model include weight ω and bias term 8. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 6, characterized in that: In step S4, model training and optimization include: using an optimization algorithm to minimize the classification error; the optimization algorithm uses a stochastic gradient descent algorithm, and the stochastic gradient descent algorithm includes: Determine the loss function: Use nonlinear logical operations to interpret the aggregated rules and derive the loss function as follows: The initial values of the learning rate and the number of training rounds are set to 0.1 and 100 respectively. For each training sample x t Calculate the loss function J t Regarding the gradient of the model parameters, the formula is: Where ω is the weight, Update the model parameters ω and bias terms by calculating the gradient After reaching the scheduled training rounds, the final evaluation results of the model are calculated; The data set is randomly shuffled and divided into 5 subsets. The i-th subset is selected as the validation set, and the other 4 subsets are used as training sets. The training set is used to train model 1 to obtain model M. i , use the validation set to evaluate the trained model and calculate the evaluation indicators; after 5 rounds of training and validation, 5 evaluation indicators are obtained; by observing the changes of these indicators on the validation set, it is determined whether the model is overfitting or underfitting.
9. The binary classification method for simulating human pairwise comparison in a decision-making process according to claim 1, characterized in that: In step S5, after obtaining the binary classification result using the trained model, the category label or the probability value belonging to each category is output.
10. A binary classification system that simulates human pairwise comparison in decision-making, characterized by: The system is used to control the binary classification method of human pairwise comparison in the simulated decision-making process described in any one of claims 1 to 9, and the system comprises: The data preprocessing module is used to clean the input data, remove noise and outliers, and perform feature normalization and standardization to make different features have the same scale; Feature extraction module, used to introduce random forest method to extract data features; The classification model building module is used to build a binary classification model by simulating the human pairwise comparison method in the decision-making process using multiplication priority; The model training and optimization module is used to train the classification model using labeled training data, evaluate the performance and generalization ability of the model using cross-validation techniques, and select the model parameters with the best performance; The classification result output module is used to input the data to be classified into the trained model to obtain the binary classification results.