Diabetes risk assessment method, terminal equipment and storage medium
Through multimodal data integration and FP-Growth algorithm mining association rules, combined with stacking algorithm fusion model, the problems of data restriction and a single data source in diabetes risk assessment are solved, and more efficient and accurate risk assessment is achieved.
Patent Information
- Application Number
- CN202410975223.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has limited data in diabetes risk assessment, insufficient utilization of a single data source, and strong subjectivity in feature extraction, resulting in low model training effect and evaluation accuracy.
Multimodal data integration is adopted, frequent item sets and association rules are mined through the FP-Growth algorithm, and multiple machine learning models are fused with the stacking algorithm to generate meta-models for diabetes risk assessment.
在数据受限情况下,充分利用多数据源信息,提高模型训练效果和评估准确性,避免主观性影响,提供更全面和客观的糖尿病风险评估。
Smart Images

Figure CN118969269A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of smart medical care, and in particular to a diabetes risk assessment method, terminal equipment and storage medium. Background Art
[0002] Diabetes is a common chronic metabolic disease, especially type 2 diabetes, which is characterized by insulin resistance and insufficient insulin secretion. With the change of lifestyle and the aging of the population, the prevalence of diabetes has increased rapidly, posing a huge challenge to global public health and economy. In the screening and management of diabetes, different types of data are often involved, including but not limited to medical images (X-rays, CT, etc.), genomic data, electronic health records (EHRs), lifestyle data (diet, exercise, etc.) and environmental data. The integration and analysis of these multimodal data help relevant personnel to fully understand the health status of patients and provide personalized treatment and intervention plans. With the development of big data technology, data mining and machine learning methods are increasingly used in medical data analysis, especially association mining technology, which can discover hidden association patterns from a large amount of complex data and provide a new perspective for disease screening and prediction. Association mining mainly includes frequent item set mining and association rule generation. These technologies have achieved remarkable results in market basket analysis, text analysis and other fields, and have gradually been applied to the analysis of medical data.
[0003] Convolutional neural networks (CNNs) have long dominated the field of artificial intelligence. Therefore, in traditional research methods, medical imaging data is often used to screen for diabetic complications. First, medical images are preprocessed, such as noise removal, image enhancement, and contrast adjustment, to improve image quality and the accuracy of subsequent analysis. Then, CNN technology is used to automatically extract features from image data and classify them. Studies have shown that image analysis methods based on deep learning show high accuracy in the detection of diabetic retinopathy. In addition to this research method, analyzing the information contained in electronic health records is also one of the traditional methods. First, through data cleaning and standardization, the data is integrated to form a structured database to solve the problems of inconsistent formats and many missing values. Then, relevant features are extracted from it, such as the patient's age, gender, BMI, blood sugar level, family history, etc., and important features related to diabetes are screened out through feature selection methods (such as LASSO regression, random forest, etc.). Finally, based on the extracted features, diabetes risk scoring models, such as Framingham risk score, are constructed. These models are usually trained and verified using traditional machine learning methods such as logistic regression.
[0004] Traditional methods require large-scale data sets to train models, which is not practical in many data-limited scenarios. The feature extraction and selection process relies on expert knowledge and is subjective. Single data has limited ability to screen for diabetes risk and lacks comprehensive use of multimodal data. Summary of the invention
[0005] The technical problem to be solved by the present invention is to provide a diabetes risk assessment method, terminal device and storage medium in view of the shortcomings of the existing technology, which can fully utilize the information of different data sources even when the data is limited, thereby improving the training effect and evaluation accuracy of the model.
[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a diabetes risk assessment method, comprising the following steps:
[0007] S1. Acquire multimodal diabetes data of a user, wherein the multimodal diabetes data includes clinical diagnosis information, medical images, gene data, and lifestyle data;
[0008] S2. extracting features of the multimodal diabetes data, and concatenating data features of different modalities by row or column to obtain a feature matrix;
[0009] S3, mining the feature matrix to obtain an association rule set;
[0010] S4. Use the features selected from the association rule set to train N different machine learning models, and concatenate the outputs of the N machine learning models to obtain a new feature vector; use the new feature vector as the input of the meta-model, train the meta-model, and obtain an evaluation model.
[0011] The specific implementation process of step S3 includes:
[0012] 1) Convert the feature matrix into a transaction database format, where each transaction in the transaction database represents a feature set of a user, calculate the occurrence frequency of each feature item in the feature set, and sort and reorganize each transaction in descending order of occurrence frequency;
[0013] 2) Insert the reorganized transactions into the FP tree. Each path in the FP tree represents a feature set of a transaction. Extract frequent itemsets from the FP tree and calculate the support of each frequent item set. If the support of a frequent item set is greater than the preset minimum support threshold, the frequent item set is retained.
[0014] 3) For each filtered frequent item set, generate all possible association rules, the antecedent and the consequent of the association rule are both true subsets of the frequent item set, and the union of the antecedent and the consequent is equal to the frequent item set;
[0015] 4) Calculate the confidence of each association rule. If the confidence is greater than the preset minimum confidence threshold, the association rule is retained.
[0016] In step S4, the specific implementation process of training the meta-model to obtain the evaluation model includes:
[0017] The unscreened features in the association rule set are used as a validation set, and the performance indicators of the N different machine learning models are evaluated using the validation set. The weight of each machine learning model is determined according to the performance indicator of each machine learning model, and each weight value is multiplied by the corresponding element in the result vector output by the meta-model to obtain the final evaluation model.
[0018] As an inventive concept, the present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the above method.
[0019] As an inventive concept, the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon; the computer program / instruction implements the steps of the above method when executed by a processor.
[0020] As an inventive concept, the present invention also provides a computer program product, including a computer program / instruction; when the computer program / instruction is executed by a processor, the steps of the above method are implemented.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] 1) The present invention integrates multimodal data, that is, combines multiple data, so that even when data is limited, it can make full use of information from different data sources, thereby improving the training effect and evaluation accuracy of the model. In addition, the FP-Growth algorithm is used to efficiently mine frequent item sets and association rules, which can discover important potential associations under limited data conditions and improve data utilization efficiency;
[0023] 2) The present invention automatically extracts features from multimodal data through machine learning algorithms, and uses FP-Growth algorithms and stacking algorithms for feature selection and model training, avoiding the influence of subjectivity in traditional methods and ensuring the objectivity and scientificity of feature selection;
[0024] 3) The present invention improves the comprehensiveness and richness of information by integrating multimodal data, and can more comprehensively assess diabetes risk. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1The principle of establishing the header table of the FP-Growth algorithm;
[0026] Figure 2 The schematic diagram for establishing the FP tree;
[0027] Figure 3 This is a flow chart of the Stacking algorithm according to an embodiment of the present invention;
[0028] Figure 4 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0029] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0030] Example 1
[0031] The embodiment of the present invention aims to improve the accuracy of diabetes risk screening by combining multimodal feature fusion and association mining technology. Through data preprocessing, feature extraction and fusion, association rule mining, model construction and other steps, a comprehensive assessment of diabetes risk is achieved.
[0032] First, it is necessary to collect multimodal data for analysis, including but not limited to clinical indicator data (blood sugar, blood pressure, etc.), genomic data, lifestyle data (eating habits, exercise frequency, etc.) and medical imaging data in the EHR system. These data will serve as the basis for subsequent feature extraction, association mining and risk assessment. The embodiment of the present invention will obtain these data from public data sets. Afterwards, the data is preprocessed to ensure the quality of the data, including the processing of missing values and outliers, the calculation of new features (average blood sugar, etc.), the alignment of time series data (number of steps, heart rate), and the unique hot encoding of individual data (gender, etc.).
[0033] After preprocessing the data set, a series of features related to diabetes risk are extracted from it, such as SNPs in genomic features, FPG, HbA1c and other indicators in EHR, and the mean and standard deviation of blood glucose measurement, and the features are spliced and transformed. If the feature dimension is too high, PCA technology is used to reduce the number of features and convert the high-dimensional feature space into a low-dimensional feature space while retaining the main variability of the data. Afterwards, based on these features, we use the FP-Growth algorithm in association mining to explore the meaningful rules between different features and diabetes risk. This algorithm is a method based on the Apriori algorithm and evolved. Its significant advantage is that it uses a data structure called FP tree to store compressed frequent item set information, and only needs to scan the database twice, thus avoiding the problem of the Apriori algorithm generating a large number of candidate item sets and the need to scan the database multiple times, significantly reducing the amount of data that the algorithm needs to process. Relatively speaking, it is very suitable for discovering frequent patterns and rules between variables in large data sets. In terms of steps, it is divided into scanning the database, setting the minimum support and minimum confidence, establishing the item header table, building the FP tree, recursively mining frequent item sets, generating strong association rules from frequent item sets, and evaluating the significance and screening of rules.
[0034] After generating association rules, we need to identify a series of feature combinations that are most important for diabetes risk assessment, such as a rule: {high blood sugar, high blood pressure}->increased risk of diabetes, with a support of 0.05 and a confidence of 0.8. This means that in the dataset, 5% of patients have both high blood sugar and high blood pressure, and 80% of these patients are diagnosed with diabetes. This rule shows that high blood sugar and high blood pressure are strong indicators of increased risk of diabetes. We can mark these two features as high priority features and include them in the feature set for subsequent model training.
[0035] In model training, the embodiment of the present invention uses three basic machine learning models (SVM, RF, K-NN) as tools for diabetes risk screening, trains the feature set identified in the association rule mining step, clarifies the target variable (diabetes diagnosis result), obtains multiple basic models, and records the prediction results of each model on the training set. Then, the outputs of these basic models are stacked into a matrix by stacking the stacking algorithm, and a meta-model (logistic regression or GBDT) is trained as input for fusion. The meta-model performs the final classification by learning these new features. After the training is completed, the average performance index of the model is calculated by verification, and the model parameters are adjusted based on the result. The optimal parameter combination is found by random search, such as the number of trees in the random forest model and the regularization parameter in the SVM model. Finally, the optimized model is deployed in the actual application environment for risk screening and prediction. Monitor the model performance regularly, and update and retrain the model according to new data.
[0036] The ultimate goal of the embodiment of the present invention is to achieve efficient association mining and characterization of multimodal big data and provide accurate diabetes risk assessment and screening results. The method flow is as follows: Figure 4 As shown:
[0037] Step 1: Collect multimodal data of diabetic patients from multiple public datasets, including clinical diagnosis information (MIMIC-III), medical imaging (NIH Chest X-ray Dataset), genetic data (1000Genomes Project), and lifestyle data (NHANES), and preprocess the data;
[0038] Step 2: Use statistical analysis or other traditional machine learning algorithms to extract features from multimodal data, and then splice the data features of different modalities by row or column to form a high-dimensional feature matrix. For example, clinical features can be represented by [age, gender, blood sugar level, blood pressure, BMI...], image features can be represented by [fat distribution, visceral fat mass...], and the feature vector after all features are spliced can be represented by [age, gender, blood sugar level, blood pressure, BMI..., fat distribution, visceral fat mass..., SNP1, SNP2..., dietary habits, exercise frequency...]. After splicing the features, use standardization to return the mean of each feature to zero and the standard deviation to one to ensure that the scale of the features is the same. After the transformation is completed, a unified feature vector is formed. The fused feature vector contains the features of different modalities and eliminates the differences between modalities through transformation technology. If the feature dimension is too high (when the number of samples in the data set is much less than the number of features, the model will become complex and prone to overfitting. Generally speaking, if the number of features is more than 1 / 10 of the number of samples, the feature dimension is considered too high. For example, if there are only 100 samples in the data set, but each sample has more than 10 features, then the dimension is too high. Similarly, if there are more than 50% zero-value features in the data set, dimensionality reduction is also required), PCA is used for dimensionality reduction;
[0039] Step 3: Use the FP-Growth algorithm to mine association rules for the features fused in the previous step. First, convert the feature vector into a transaction database format suitable for processing by the FP-Growth algorithm. Each transaction represents a feature set of a patient, in the following format - transaction x: {feature A, feature B, feature C...}. After that, scan the database, calculate the support (frequency of occurrence) of each feature item, and then sort and reorganize each transaction from high to low according to the support. Then, insert each transaction into the FP tree in turn (the FP tree is initially a root node), and each path of the tree represents a feature set of a transaction. After constructing the FP tree, extract frequent item sets from the FP tree. The FP-Growth algorithm recursively constructs a conditional FP tree (see: Yuan Ming. Application of improved FP-Growth algorithm in examination results analysis [J]. Information Technology and Informatization, 2021, (06): 53-55. [2] Zeng Y, Yin S, Liu J, et al. Research of improved FP-growth algorithm for mining association rules mining [J]. Scientific Programming, 2015, 2015 (1): 910281.) to mine frequent itemsets. For each frequent item set, its support is calculated. If the support is greater than the preset minimum support threshold, the frequent item set is retained. For each filtered frequent item set, all possible association rules are generated. The antecedent and consequent of the association rule (for example, A->B, A is the antecedent, B is the consequent, or A->BC or AB->C, etc., the antecedent is before the arrow, and the consequent is after the arrow. In association rule mining, items usually refer to elements in a transaction database.) are both true subsets of the frequent item set, and the union of the antecedent and consequent is equal to the frequent item set. Finally, the confidence is calculated for each candidate rule (that is, all possible association rules generated above). The confidence represents the probability of the consequent occurring when the antecedent occurs. The formula is confidence (A->B) = support (A, B) / support (A). If the confidence is greater than the preset minimum confidence threshold, the rule is retained. After obtaining the rules, we can find the potential associations between diabetes-related risk factors, and the features involved in the association rules will also be used for our subsequent model training;
[0040] In the embodiment of the present invention, when setting the minimum support threshold and the minimum confidence threshold, it is necessary to determine them through trial and error, that is, to adjust them through experiments and verification. First, it is necessary to determine the minimum support threshold, which is generally set to 0.1, run the FP-Growth algorithm, and observe the number and quality of the frequent item sets mined. If the number of frequent item sets is too large, it means that the threshold is set too low and the threshold needs to be increased; if the number of frequent item sets is too small, the threshold needs to be lowered. Generally speaking, in large data sets, the number of frequent item sets accounts for about 1% of all item sets. After determining the support, the minimum confidence threshold is set, which is generally set to 0.7 first, so that the rules have high reliability, and then the confidence threshold is adjusted through experiments and verification. If too many rules are mined, the threshold can be increased; if there are too few rules, the threshold can be appropriately lowered. The number of rules is generally controlled to be 20-50.
[0041] Step 4: Use the features selected from the association rule set to train three basic machine learning models (SVM, RF, K-NN) to obtain three basic models, and then use the stacking algorithm to stack the outputs of the three basic models (i.e., prediction probabilities) together to form a new feature vector as the input of the subsequent meta-model, thereby effectively integrating the outputs of the basic models and improving the overall prediction performance. The meta-model in the present invention uses a logistic regression model. Finally, according to the performance of the basic model on the validation set, the weight of each model is determined, and the elements in the output result vector of the meta-model are weighted averaged. The result after weighted average is used as the output of the final diabetes risk assessment model. On this basis, we train the meta-model and strive to improve the accuracy;
[0042] In the present invention, the meta-model will output a feature vector (p4, p5, p6), which is the result of processing (p1, p2, p3). p1, p2, p3 are the predicted probabilities output by the three basic models. Afterwards, the performance of the three basic models is evaluated on the validation set, and their performance indicators (such as accuracy, F1 score, etc.) are calculated. Based on these indicators, the weight of each model is determined. For example, if the weights are w1, w2, and w3 respectively, the final output is ((w1×p4)+(w2×p5)+(w3×p6)) / (w1+w2+w3).
[0043] Step 5: After training the model in step 4, evaluate its performance based on the validation set. After completing the first round of evaluation, use random search to adjust the hyperparameters to further improve the performance of the model. In the entire process of adjusting the hyperparameters, first determine the hyperparameters that need to be adjusted in the model and their related search spaces, such as the number of decision trees in the RF model (10, 50...), the maximum depth (10, 20...), the number of neighbors k in the K-NN model (3, 5, 7, etc.), the weight type (such as equal weight, distance weight, etc.), etc.; then, determine the number of iterations of random search, that is, the number of times the hyperparameter combination is randomly selected. Generally speaking, the more iterations, the greater the probability of finding the best hyperparameter combination, but the computational cost will also increase accordingly. The common number of iterations ranges from 100 to 1000. In each iteration, a hyperparameter combination is randomly selected. The following is an example process of randomly selecting a hyperparameter combination: the number of decision trees for the random forest model is selected as 50, the maximum depth is 20, and the minimum number of sample splits is 5. Select the kernel function type of the SVM model as RBF kernel, the penalty parameter C as 1, and the gamma parameter as 0.1. Select the number of neighbors K of the K-NN model as 5, and the weight type as distance weight.
[0044] For each randomly selected hyperparameter combination, the present invention uses a validation set to evaluate the performance of the model. Common performance indicators include accuracy, recall, F1 score, etc., and the performance indicators corresponding to each hyperparameter combination are recorded. After all iterations are completed, the performance indicators of each hyperparameter combination are compared, and the hyperparameter combination with the best performance indicator is selected. If necessary, the best hyperparameter combination can be used to retrain the basic model and metamodel to ensure the best performance of the final model. Finally, the optimized model is deployed in a practical environment for application to provide accurate diabetes risk assessment.
[0045] Table 1 Source data of users in the embodiment of the present invention
[0046]
[0047] Table 1 above shows the source data of some users, a total of 2,000 records. The clinical, imaging, genetic and lifestyle characteristics of each user are extracted, and the features of different modalities are spliced into a unified feature vector and standardized.
[0048] Example feature vector: [45,M,120,130 / 80,25,0.2,0.15,AA,CC,1,3], convert the feature vector into transaction database format and perform association rule mining. The experimental results are as follows:
[0049] Example transaction:
[0050] Transaction 1:{Age=45, Gender=M, Blood Glucose=120, BMI=25, FatDistribution=0.2, SN P1=AA, Diet=1}
[0051] Example rule:
[0052] Rule 1: {Age = 45, Blood Glucose = 120} -> {BMI = 25};
[0053] Rule 2: {Gender = M, SNP1 = AA} -> {Blood Glucose = 120};
[0054] According to the rules, features such as Gender, SNP1, and BMI can be screened out, and the three basic models are trained, and the prediction probabilities of the basic models are spliced to form a new feature vector as the input of the meta-model. The present invention uses logistic regression as the meta-model. After the three basic models are trained, the three basic models are evaluated using the validation set. The prediction accuracy of the three basic models is SVM: 85.43%, RF: 87.24%, and K-NN: 82.58%, respectively. Therefore, the model finally uses 85.43, 87.24, and 82.58 as weights, and performs weighted average on the output vector of the meta-model to obtain the final value. On this basis, the meta-model is trained again, and its final evaluation accuracy on the validation set is 90.92%, which exceeds any basic model, so the stacking algorithm is conducive to optimizing the model.
[0055] The multimodal big data association mining and characterization method of the embodiment of the present invention integrates multimodal data such as clinical, imaging, genetic and lifestyle data, uses the FP-Growth algorithm to efficiently mine association rules and the stacking algorithm to integrate multiple basic models, thereby improving the prediction performance and generalization ability of the risk assessment model and the accuracy of diabetes risk screening. Compared with traditional methods, the method of the embodiment of the present invention is more comprehensive, efficient and flexible, and can provide strong technical support for early prevention of diabetes and personalized health management.
[0056] The embodiments of the present invention solve the problems of single information, low data processing efficiency and insufficient model accuracy in existing diabetes risk screening methods. Existing technologies usually rely on a single data source, resulting in incomplete risk assessment results. At the same time, traditional data mining algorithms are inefficient when processing large-scale and diversified data, making it difficult to discover potential associations. The predictive performance of a single model is also limited, and it is impossible to fully utilize information from multiple data sources. The embodiments of the present invention provide a more comprehensive, accurate and efficient diabetes risk assessment method by integrating multimodal data, using the FP-Growth algorithm for efficient association rule mining, and using the stacking algorithm to fuse multiple basic models, which significantly improves the accuracy and reliability of screening.
[0057] Example 2
[0058] Embodiment 2 of the present invention provides a terminal device corresponding to the above-mentioned embodiment 1. The terminal device may be a processing device for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of the above-mentioned embodiment.
[0059] The terminal device of this embodiment includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program in the memory to implement the steps of the method in the above-mentioned embodiment 1.
[0060] In some implementations, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.
[0061] In some other implementations, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors of various types, which are not limited herein.
[0062] Example 3
[0063] Embodiment 3 of the present invention provides a computer-readable storage medium corresponding to the above embodiment 1, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method of the above embodiment 1 are implemented.
[0064] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.
[0065] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.
[0066] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0067] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0068] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.
[0069] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
Claims
1. A diabetes risk assessment method, characterized in that: The following steps are involved: S1. Acquire multimodal diabetes data of a user, wherein the multimodal diabetes data includes clinical diagnosis information, medical images, gene data, and lifestyle data; S2. extracting features of the multimodal diabetes data, and concatenating data features of different modalities by row or column to obtain a feature matrix; S3, mining the feature matrix to obtain an association rule set; S4. Use the features selected from the association rule set to train N different machine learning models, and concatenate the outputs of the N machine learning models to obtain a new feature vector; use the new feature vector as the input of the meta-model, train the meta-model, and obtain an evaluation model.
2. The diabetes risk assessment method according to claim 1, characterized in that: The specific implementation process of step S3 includes: 1) Convert the feature matrix into a transaction database format, where each transaction in the transaction database represents a feature set of a user, calculate the occurrence frequency of each feature item in the feature set, and sort and reorganize each transaction in descending order of occurrence frequency; 2) Insert the reorganized transactions into the FP tree. Each path in the FP tree represents a feature set of a transaction. Extract frequent itemsets from the FP tree and calculate the support of each frequent item set. If the support of a frequent item set is greater than the preset minimum support threshold, the frequent item set is retained. 3) For each filtered frequent item set, generate all possible association rules, the antecedent and the consequent of the association rule are both true subsets of the frequent item set, and the union of the antecedent and the consequent is equal to the frequent item set; 4) Calculate the confidence of each association rule. If the confidence is greater than the preset minimum confidence threshold, the association rule is retained.
3. The diabetes risk assessment method according to claim 1, characterized in that: In step S4, the specific implementation process of training the meta-model to obtain the evaluation model includes: The unscreened features in the association rule set are used as a validation set, and the performance indicators of the N different machine learning models are evaluated using the validation set. The weight of each machine learning model is determined according to the performance indicator of each machine learning model, and each weight value is multiplied by the corresponding element in the result vector output by the meta-model to obtain the final evaluation model.
4. A terminal device comprising a memory, a processor and a computer program stored in the memory; characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 3.
5. A computer-readable storage medium having a computer program / instruction stored thereon; characterized in that: When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 3 are implemented.
6. A computer program product comprising a computer program / instructions; characterized in that: When the computer program / instructions are executed by a processor, the steps of the method of any one of claims 1 to 3 are implemented.
Citation Information
Patent Citations
Diabetes risk early warning method based on big data analysis
CN117253614A
Multi-modal prediction model construction method and system for analyzing liver cancer recurrence data
CN117612711A
Chronic disease health monitoring and early warning system and method based on data analysis
CN118136270A
Cited By
Squeezing risk assessment device and method
CN120809241A
Device and method for assessing crush risk
CN120809241B
Multi-mode diabetes risk prediction method based on combination of traditional Chinese medicine and western medicine
CN121171601A
