A method and system for cardiovascular disease risk assessment based on machine learning models

Abstract

This invention relates to a method and system for cardiovascular disease risk assessment based on machine learning models. By utilizing machine learning models to focus on predicting the risk of heart failure, it accurately distinguishes between healthy individuals and those with cardiovascular disease, emphasizing the differentiation of heart failure and other common cardiovascular diseases, and deeply exploring diagnostic indicators, thus opening new avenues for cardiovascular diagnosis and treatment. It hierarchically utilizes features closely related to blood cells, blood lipids, and blood glucose, and constructs predictive models using multiple machine learning models, laying a solid foundation for further in-depth research. This invention fully utilizes diagnostic indicators and multiple machine learning models, making full use of the predictive results and differentiated characteristics of different machine learning models to achieve efficient and accurate 5-category risk assessment.

Description

[Technical Field]

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a method and system for cardiovascular disease risk assessment based on machine learning models. [Background Technology]

[0002] As the aging population intensifies, chronic diseases have become the leading cause of death globally. Cardiovascular disease, in particular, is a major health concern for the elderly, and society faces immense pressure from its continued rise. Cardiovascular diseases refer to those affecting the function of the heart and blood vessels, including common conditions such as hypertension, myocardial infarction, coronary heart disease, heart failure, heart disease, and arrhythmias. Currently, there is no effective management model for chronic diseases. Cardiovascular disease, also known as cardiovascular and cerebrovascular disease, has the highest mortality rate, exceeding that of cancer and other diseases.

[0003] With the development of big data and artificial intelligence technologies, combined with intelligent models and the mature application of intelligent wearable early warning devices, there is significant room for advancement in research on the comprehensive and precise assessment, regular monitoring, effective intervention and prevention, and early clinical diagnosis and treatment of cardiovascular and cerebrovascular disease risk factors. With the development of the computer field, intelligent medical diagnostic systems have emerged. The knowledge base in these systems stores diagnostic rules for diseases. These rules primarily describe the manifestations and characteristics of diseases, representing the causal relationship between diseases and symptoms. Early knowledge bases were organized entirely based on the thought processes of clinicians. Manually compiling the knowledge base based on doctors' experience was tedious, time-consuming, and labor-intensive. Furthermore, the generation of rules had certain human limitations, insufficiently considering the diversity, variability, and uncertainty of diseases, and potentially overlooking certain special rules.

[0004] While methods for generating rule bases using machine learning are emerging, they still lag behind in terms of technological maturity and rule rationality, resulting in low diagnostic accuracy in practical applications and significant room for improvement. Domestic and international cardiovascular disease risk prediction and assessment are primarily based on predictive variables such as blood pressure, body mass index, blood glucose, total cholesterol, smoking, medical history, and family history. Traditional cardiovascular disease risk assessment models suffer from insufficient feature representation, weak model generalization ability, and low accuracy and reliability. Optimization of traditional models also suffers from low search efficiency, susceptibility to local optima, and reliance on human experience for parameter adjustments. Machine learning-based cardiovascular disease risk assessment systems represent an innovative medical technology. They utilize machine learning algorithms to deeply mine and learn from medical data, constructing cardiovascular disease risk assessment models capable of accurate prediction and assessment of cardiovascular disease risk. Currently, many studies use neural networks, even deep neural networks, for prediction and assessment. Neural networks typically consist of multiple layers, including input, hidden, and output layers, each composed of multiple neurons. Traditional machine learning models, however, have relatively simple structures; for example, linear models have only simple parameters, and decision trees have branching structures, but unlike neural networks, they lack multi-layered structures. Neural networks are more complex than traditional machine learning models, and their training is very difficult, making it nearly impossible to achieve a valuable training objective. Machine learning models, on the other hand, are relatively simple to use, train quickly, and accumulate a wealth of training experience. How to fully leverage the advantages of machine learning models to achieve intelligent upgrades in risk assessment is a problem to be solved. Based on the above issues, this invention fully utilizes diagnostic disease indicators and multiple machine learning models, making full use of the prediction results and differentiated characteristics of different machine learning models to achieve efficient and accurate 5-category risk assessment. [Summary of the Invention]

[0005] To address the aforementioned problems in the prior art, this invention proposes a method and system for cardiovascular disease risk assessment based on a machine learning model, the method comprising:

[0006] Step S1: Construct I machine learning models as prediction models, and select N user features as candidate input features;

[0007] Step S2: Select N1 user features from the candidate input features as the first input features of the prediction model;

[0008] Step S3: Train and validate the prediction model based on N1 user features of the sample, and set evaluation parameters to evaluate the prediction model;

[0009] Step S4: Construct a discrimination sequence associated with each pair of categories based on the evaluation parameters; the discrimination sequence contains a prediction model identifier and is sorted according to the ability of the prediction model to distinguish between the two categories;

[0010] Step S5: Obtain I classification results R for the N1 user features of the user to be evaluated, corresponding to I prediction models. i =(r i,k ); where: r i,k It is the k-th element of the classification result of the i-th prediction model; determine the maximum element value max(r) for each of the I classification results. i,k If the maximum value of the element in the classification results of I prediction models all points to the same category, then the mean of the I classification results is used as the evaluation result; if M1 prediction models greater than or equal to PER1 point to the same category k1, the category pointed to by the maximum value of the element in the classification results of the prediction models that do not point to the same category is determined and called category k2. The identification sequence corresponding to the classification group (k1, k2) is obtained, identification is performed based on the identification sequence, and the identification result is used as the evaluation result.

[0011] Furthermore, I=5, meaning there are 5 different machine learning models.

[0012] Furthermore, the five machine learning models are LR, GB, NNK, GNB, and SVM machine learning models.

[0013] Furthermore, N = 29.

[0014] Furthermore, N1 = 20.

[0015] Furthermore, the N1 user features are arranged sequentially to form an N-dimensional vector, which serves as the first input feature of the machine learning model.

[0016] Furthermore, PER1 = 50%.

[0017] Furthermore, the identification based on the identification sequence is specifically as follows: at this time, the largest element value of the M1 prediction results points to the same category k1, and the classification result of the I-M1 prediction model that is last sorted in the identification sequence is deleted; if all the classification results that have not been deleted point to the same category, the mean of all the classification results that have not been deleted is used as the evaluation result; otherwise, I is set to M1 and the process returns to step S5.

[0018] Furthermore, before training the prediction model, the sample data used for training is preprocessed to remove all samples containing missing values.

[0019] A cardiovascular disease risk assessment system based on a machine learning model is provided, wherein the machine learning model-based cardiovascular disease risk assessment system is used to implement the aforementioned machine learning model-based cardiovascular disease risk assessment method.

[0020] The beneficial effects of this invention include:

[0021] (1) Make full use of diagnostic disease indicators and multiple machine learning models, and make full use of the prediction results and their differences of different machine learning models to achieve efficient and accurate 5-category risk assessment.

[0022] (2) For different users' prediction results, two classifications are used to construct identification sequences and identification matrices to select multiple prediction models, which can make full use of the classification results of different prediction models. When the classification results are inconsistent but the differences between different prediction models are not obvious, by constructing an identification matrix, the evaluation of prediction models in all dimensions can be fully utilized to find the prediction model with the most obvious difference to participate in the prediction evaluation, so as to comprehensively use the learning results of multiple prediction models. By using diagnostic disease indicators in a hierarchical manner, the classification confusion caused by input parameters to the prediction model can be eliminated, so that multiple prediction models can achieve rapid consistency under the condition of basic consistency. At the same time, the method is also highly scalable. [Attached Image Description]

[0023] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to unduly limit the invention. In the drawings:

[0024] Figure 1 A schematic diagram of the cardiovascular disease risk assessment method based on a machine learning model provided by the present invention.

[0025] Figure 2 shows the predicted ROC curves of the five machine learning models provided in this invention. Specifically: Figure 2(A): Patients with coronary atherosclerotic heart disease and healthy individuals; Figure 2(B): Patients with unstable angina and healthy individuals; Figure 2(C): Patients with heart failure and healthy individuals; Figure 2(D): ROC curves of the five machine learning models in predicting heart failure and other cardiovascular diseases.

[0026] Figure 3 This is a schematic diagram illustrating the assessment of predicting coronary atherosclerotic heart disease in healthy individuals, as provided by the present invention.

[0027] Figure 4 This is a schematic diagram illustrating the assessment of predicting unstable angina in healthy individuals, provided by the present invention.

[0028] Figure 5 This is a schematic diagram illustrating the assessment of predicting heart failure in healthy individuals, provided by the present invention.

[0029] Figure 6 This is a schematic diagram illustrating the assessment for predicting heart failure and other cardiovascular diseases provided by the present invention.

Detailed Implementation Methods

[0030] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0031] This invention proposes a method and system for cardiovascular disease risk assessment based on a machine learning model, as shown in the attached figure. Figure 1 As shown, the method includes the following steps:

[0032] Step S1: Construct I machine learning models as prediction models, and select N user features as candidate input features;

[0033] Preferably: I=5, the 5 machine learning models are 5 different machine learning models, namely LR, GB, NNK, GNB and SVM machine learning models; the N user features are features closely related to blood cells, blood lipids and blood glucose;

[0034] Preferred: N=29; 29 user characteristics include sex, age, absolute basophil count (BAS#), basophil percentage (BAS%), absolute eosinophil count (EOS#), eosinophil percentage (EOS%), hematocrit (HCT), hemoglobin (HGB), absolute lymphocyte count (LYM#), lymphocyte percentage (LYM%), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular volume (MCV), and monocyte count. The following parameters were measured: mean platelet count (MON#), monocyte percentage (MON%), mean platelet volume (MPV), absolute neutrophil count (NEU#), neutrophil percentage (NEU%), plateletcrit (PCT), platelet distribution width (PDW), platelet count (PLT), red blood cell count (RBC), red blood cell distribution width (RDW), white blood cell count (WBC), glucose (GLU), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), and triglycerides (TG).

[0035] Step S2: Select N1 user features from the candidate input features as the first input features of the prediction model; specifically: assign a SHAP value to each user feature, the SHAP value is used to indicate the impact of the user feature on the prediction model; select N1 user features from the candidate input features as the first input features of the prediction model based on the SHAP value;

[0036] Preferred: Arrange the N1 user features sequentially to form an N-dimensional vector as the first input feature;

[0037] Preferably, the prediction model's classification results include categories 1-5, namely coronary atherosclerotic heart disease, unstable angina, heart failure, other cardiovascular diseases, and health. The corresponding output of the prediction model is a 5-tuple vector, where each element corresponds to one of the five classification results. The sum of all elements in the output vector is 1, and the value of each element indicates the confidence level of the prediction model for its corresponding classification result. For example, (0.1, 0.2, 0.4, 0, 0) indicates confidence levels of 0.1, 0.2, and 0.4 for coronary atherosclerotic heart disease, heart failure, and unstable angina, respectively, with the highest confidence level for unstable angina. However, in reality, an expression of 0.4 does not provide reliable information for the assessment and has limited reference value.

[0038] Preferably, the SHAP value of the user features is calculated by the "shap python" package (version 0.41.0). For the 5 prediction models, N1=20 is set, and the top 20 key user features obtained by the SHAP algorithm are used as the first input features. In this algorithm, the "Counter" function is used to count the feature word frequency. The word frequency is calculated by flattening the data, and an importance score is constructed based on the feature importance ranking. The comprehensive score is generated by combining the word frequency and importance to identify key features.

[0039] Alternative: The top N1 key user features are determined based on user feedback as the first input feature; users can select key user features based on expert experience;

[0040] Step S3: Train and validate the prediction model based on N1 user features of the sample, and set one or more evaluation parameters to evaluate the prediction model; specifically: train and validate the prediction model based on N1 user features of the sample, and evaluate the performance of each prediction model by determining the evaluation parameters based on the area under the ROC curve (AUC).

[0041] Preferred method: Before training the prediction model, preprocess the sample data used for training and remove all samples containing missing values.

[0042] Preferred approach: For each prediction model, the data of healthy individuals and patients with diseases are randomly divided into training set and validation set. The "Sample" function in the "Pandas" package is used to randomly divide the data of all cases into a training set (70%) for model development and a test set (30%) for performance evaluation. The training set is used to train the model, and the test set is used to evaluate the performance of the model.

[0043] Preferably, the evaluation parameters include the area under the ROC curve (AUC), accuracy (Acc), sensitivity (Sn), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV), Matthews correlation coefficient (MCC), and / or F1 score. The model performance is evaluated based on these eight evaluation parameters. The evaluation parameters are determined using the following formulas (1) to (7). Wherein, TP, TN, FP, and FN represent true positive, true negative, false positive, and false negative, respectively.

[0044] ACC=(TP+TN) / (TP+TN+FP+FN) (1);

[0045] Sn=TP / (TP+FN) (2);

[0046] Sp=TN / (TN+FP) (3);

[0047] PPV = TP / (TP+FP) (4);

[0048] NPV = TN / (TN+FN) (5);

[0049]

[0050] F1=2TP / (2TP+FN+FP) (7);

[0051] The eight evaluation parameters mentioned above are used in different scenarios. For example, Accuracy is effective when the data is balanced, but may be misleading in imbalanced cases; AUC is more stable, especially in binary classification problems where sensitivity and specificity are for different classes, while PPV and NPV are probabilities of the true class derived from the prediction results; MCC considers all four elements of the confusion matrix, so it is more meaningful when the classes are imbalanced; F1 is suitable when precision and recall are equally important. Therefore, for prediction results of different users, using two classifications to construct classification groups based on AUC to select multiple prediction models can make full use of the classification results of different prediction models.

[0052] Step S4: Construct a discrimination sequence associated with each pair of categories based on the evaluation parameters; the discrimination sequence contains a prediction model identifier and is sorted according to the ability of the prediction model to distinguish between the two categories;

[0053] The discrimination sequence is specifically constructed based on AUC. After training the five prediction models, the evaluation data are shown in Figures 2-6. As can be seen from the figures, the GB model exhibits better overall performance in predicting patients with coronary atherosclerotic heart disease and healthy individuals, with an AUC of 0.97 (95% CI: 0.93-0.99). The GNB model shows the best overall performance in differentiating between patients with unstable angina and healthy individuals, with an AUC of 0.94 (95% CI: 0.89-0.99). In distinguishing between patients with heart failure and healthy individuals, the LR and GNB models have good overall performance, with corresponding AUCs of 0.99 (95% CI: 0.96-1.00) and 0.99 (95% CI: 0.97-1.00), respectively. Similarly, in the construction of models to distinguish heart failure from other cardiovascular diseases, the GB model showed excellent overall performance with an AUC of 0.81 (95% CI: 0.69-0.92); the AUCs of the LR, GNB, NN, and SVM models were 0.76 (95% CI: 0.59-0.91), 0.76 (95% CI: 0.61-0.90), 0.76 (95% CI: 0.58-0.90), and 0.76 (95% CI: 0.59-0.89), respectively; the identification sequences for classification groups (3,4) were (GB (0.81), LR (0.76), GNB (0.76), NN (0.76), SVM (0.76)); the construction of identification sequences for other classification groups was similar.

[0054] Alternatively: Step S4 specifically involves: constructing a discrimination matrix for each of the two categories based on all evaluation parameters; the element values ​​m in the discrimination matrix... i,j The parameter value of the j-th evaluation parameter of prediction model i is indicated. When the classification results are inconsistent but the differences between different prediction models are not obvious, by constructing an identification matrix, the evaluation of the prediction model in all dimensions can be fully utilized to find the prediction model with the most obvious difference to participate in the prediction evaluation, so as to comprehensively use the learning results of multiple prediction models. At the same time, the method is also highly scalable.

[0055] Step S5: Obtain I classification results R for the N1 user features of the user to be evaluated, corresponding to I prediction models. i =(r i,k ); where: r i,k It is the k-th element of the classification result of the i-th prediction model; determine the maximum element value max(r) for each of the I classification results. i,kIf the maximum value of the element in the classification results of I prediction models all points to the same category, then the mean of the I classification results is used as the evaluation result; if M1 prediction models greater than or equal to PER1 point to the same category k1, the category pointed to by the maximum value of the element in the classification results of the prediction models that do not point to the same category is determined, called category k2, and the identification sequence corresponding to the classification group (k1, k2) is obtained. Identification is performed based on the identification sequence, and the identification result is used as the evaluation result; if at most one prediction result points to the same category (all classification results point to different categories), then feedback is given;

[0056] Preferred method: In the initial state, set I=5 and input the user features to be evaluated into 5 prediction models to obtain the classification results for each prediction model;

[0057] Preferred: PER1 is a preset percentage, for example: PER1 = 50%;

[0058] Alternative: If the maximum value of the element in the classification results of I prediction models all points to the same category, then the classification result corresponding to the prediction model with the largest AUC value shall be used as the evaluation result.

[0059] Preferred method: When there are multiple categories k2, select the category with the most points to by the maximum value as category k2; when multiple categories have the same number of points to by the maximum value, randomly select one as category k2.

[0060] The feedback process specifically involves: confirming that the intelligent assessment is invalid; and indicating that there may be an anomaly in user feature collection.

[0061] The identification based on the identification sequence is as follows: at this time, the largest element value of M1 prediction results points to the same category k1, and the classification result of the I-M1 prediction model at the end of the sorting in the identification sequence is deleted; if all the remaining classification results point to the same category, the mean of all the remaining classification results is used as the evaluation result; otherwise, I is set to M1 and the process returns to step S5.

[0062] Replaceable: If all classification results that are not deleted point to the same category, then the classification result corresponding to the prediction model with the largest AUC value is used as the evaluation result;

[0063] Alternative: If all the classification results that have not been deleted do not point to the same category, then the classification result with the highest F1 score in the undeleted prediction model when identifying the classification group (k1, k2) will be used as the evaluation result.

[0064] Replaceable: If all the classification results that have not been deleted do not point to the same category, then obtain the classification results corresponding to the highest F1 score and the highest MCC coefficient when identifying the classification group (k1, k2) in the undeleted prediction model; in the classification results, r i,k1 and r i,k2 The classification result with the largest difference is used as the evaluation result;

[0065] The alternative method corresponding to step S4 is as follows: Step S1 specifically involves obtaining I classification results R corresponding to I prediction models for the user features to be evaluated. i =(r i,k ); Determine the maximum element value max(r) for each of the I classification results. i,k If the maximum value of the element in the classification results of more than or equal to I0% of the prediction models points to the same category, then the mean of the I classification results is used as the evaluation result; if the maximum value of all classification results points to different categories, then feedback is given; otherwise, the two categories that the maximum element value of the prediction result points to the most are determined and called k1 and k2 respectively, the identification matrix corresponding to the classification group (k1, k2) is obtained, identification is performed based on the identification matrix, and the identification result is used as the evaluation result.

[0066] The identification based on the identification matrix specifically involves: determining the prediction model with the largest difference in evaluation ability based on the evaluation parameters, called the difference prediction model; inserting shadow classification results based on this difference prediction model; and determining the identification result based on the classification results and shadow classification results. This includes the following steps:

[0067] Step S5A1: Take the average of the I classification results as the first shadow result;

[0068] Preferably, the initial value of I is set to 5; that is, this invention only illustrates the case of 5 prediction models, and the number of prediction models can be expanded as needed, preferably an odd number.

[0069] Step S5A2: For any row i in the identification matrix, determine any two column elements m. i,j The two elements with the largest difference between them; determine the row i containing the two elements with the largest difference in all rows and their corresponding prediction model identifiers, which are called difference prediction models; delete the columns containing the two elements in the identification matrix; in this way, the same row and its corresponding prediction model identifiers will not be selected again due to duplicate evaluation factors;

[0070] Step S5A3: Assign a first weight to the classification result of the difference prediction model, and assign a second weight to other classification results and the first shadow result; wherein: the first weight is greater than the second weight;

[0071] Preferred: The first weight is equal to twice the second weight;

[0072] Step S5A4: Based on the first weight and the second weight, obtain the weighted value of the first shadow result and other classification results as the second shadow result;

[0073] Step S5A5: Determine if more than 50% of the I+2 classification results have the maximum value of the element pointing to the same category. If so, use the mean of the I+2 classification results as the evaluation result; otherwise, set I = I+2 and return to step S5A1.

[0074] Alternatively: Step S5 specifically involves obtaining I first classification results R1 corresponding to I prediction models for the user features to be evaluated. i =(r1) i,k If more than 50% of the first classification results have the same maximum value, then the average of the I first classification results is used as the evaluation result; otherwise, proceed to the next step S6.

[0075] Step S6: Obtain I second classification results R2 corresponding to I prediction models for N2 user features of the user to be evaluated. i =(r2) i,k If the maximum value of the element in more than 50% of the second classification results of the prediction models points to the same category, then the mean of the I second classification results is used as the evaluation result; otherwise, proceed to step S7.

[0076] Alternative: If the maximum value of the element in the second classification result of more than 50% of the prediction models points to the same category, then the average of I first classification results and I second classification results is used as the evaluation result;

[0077] Preferred method: Select N2 user features from N1 user features based on SHAP values; the N2 user features are the more important N2 user features among the N1 user feature values; obtain the N2 user features of the user to be evaluated to construct the input of the prediction model; fill the parts of the input that are not the N2 user features with default values; input the input data filled with default values ​​into the prediction model respectively to obtain I second classification results corresponding to I prediction models for the N2 user features; that is, when prediction model divergence occurs, simply processing the classification results by evaluating parameters still cannot bypass the internal logic of the prediction model, and some of the N2 user features are factors that cause this divergence and are not helpful in discovering the correct classification result. Especially when divergence occurs between prediction models, too many user features will cause model "confusion". Therefore, evaluation is carried out by eliminating confounding feature factors.

[0078] Step S7: Combine the first classification result and the second classification result into a classification result. If the maximum value of the element in the second classification result of more than 50% of the prediction models points to the same category, then the mean of the 10 classification results is used as the evaluation result; otherwise, feedback is provided.

[0079] Based on the same inventive concept, the present invention also provides a cardiovascular disease risk assessment system based on a machine learning model, the system being used to perform the above-mentioned cardiovascular disease risk assessment method based on a machine learning model.

[0080] Based on the same inventive concept, the present invention also provides a cardiovascular disease risk assessment server based on a machine learning model, the server being used to perform the above-mentioned cardiovascular disease risk assessment method based on a machine learning model.

[0081] A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of programming language, including assembly or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but does not necessarily, correspond to a file in a file system. A program can be stored as part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to said program, or in multiple co-located files (e.g., a file storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on a single computer or on multiple computers located at a single site or distributed across multiple sites and interconnected by a communications network.

[0082] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for cardiovascular disease risk assessment based on a machine learning model, characterized in that, The method comprises: Step S1: constructing an I machine learning model as a prediction model, and selecting N user features as candidate input features; Step S2: selecting N1 user features from the candidate input features as first input features of the prediction model; Step S3: training and verifying the prediction model based on the N1 user features of the sample, and setting an evaluation parameter to evaluate the prediction model; Step S4: constructing a discrimination sequence associated with each of two categories based on the evaluation parameter; the discrimination sequence contains a prediction model identifier, and is sorted according to the recognition ability of the prediction model for distinguishing between the two categories; Step S5: obtaining I classification results corresponding to I prediction models for N1 user features of the user to be evaluated i = (r i,k ); wherein: r i,k is the kth element of the ith prediction model classification result; determining the classification pointed to by the maximum element value max(r i,k ) of each of the I classification results, if the maximum element values in the classification results of the I prediction models all point to the same classification, then taking the average of the I classification results as the evaluation result; if M1 prediction models greater than or equal to PER1 point to the same classification k1, determining the classification pointed to by the maximum element value in the classification result of the prediction model not pointing to the same classification, referred to as classification k2, obtaining the recognition sequence corresponding to the classification group (k1, k2), performing recognition based on the recognition sequence, and taking the recognition result as the evaluation result; wherein: PER1 is a preset percentage number; The recognition based on the discrimination sequence is specifically: at this time, the maximum element values of M1 prediction results point to the same category k1, and the classification results of I-M1 prediction models in the discrimination sequence are deleted; if all the classification results that are not deleted point to the same category, the average value of all the classification results that are not deleted is taken as the evaluation result; otherwise, I is set to M1 and the step S5 is returned. 2.The method of claim 1, wherein, I=5, and the five machine learning models are five different machine learning models. 3.The cardiovascular disease risk assessment method based on a machine learning model according to claim 2, characterized in that, The five machine learning models are LR, GB, NNK, GNB, and SVM machine learning models. 4.The method of claim 3, wherein, N＝29。 5.The method of claim 4, wherein, N1＝20。 6.The method of assessing cardiovascular disease risk based on a machine learning model according to claim 5, wherein, The N1 user features are arranged in sequence to form an N-dimensional vector as the first input features of the machine learning model. 7.The cardiovascular disease risk assessment method based on a machine learning model according to claim 6, wherein, PER1=50%. 8.The cardiovascular disease risk assessment method based on a machine learning model according to claim 7, wherein, The recognition based on the discrimination sequence is specifically: at this time, the maximum element values of M1 prediction results point to the same category k1, and the classification results of I-M1 prediction models in the discrimination sequence are deleted; if all the classification results that are not deleted point to the same category, the average value of all the classification results that are not deleted is taken as the evaluation result; otherwise, I is set to M1 and the step S5 is returned. 9.The cardiovascular disease risk assessment method based on a machine learning model according to claim 8, wherein, Before the prediction model is trained, the sample data used for training is preprocessed, and all samples containing missing values are deleted. 10.A cardiovascular disease risk assessment system based on a machine learning model, characterized in that, The cardiovascular disease risk assessment system based on the machine learning model is used to implement the cardiovascular disease risk assessment method based on the machine learning model in any one of claims 1-9.

Images