Chronic disease risk prediction model training method and system

Abstract

The present invention discloses a method and system for training a chronic disease risk prediction model, the method comprising: collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate a preliminary data feature set of multiple modalities; based on an ensemble learning algorithm, performing multi-level feature selection on the preliminary data feature set to screen key features whose impact on chronic disease risk reaches a preset standard; utilizing an attention mechanism to weight each key feature to generate a weighted feature; generating feature subsets of different levels based on the weighted feature, and utilizing each feature subset to perform hierarchical training on the to-be-trained model, so that the model gradually learns the impact of different features on chronic disease risk prediction, thereby obtaining a trained chronic disease risk prediction model. Utilizing the embodiments of the present invention, it is possible to integrate multimodal data, implement effective feature selection and weighting, and realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and in particular to a chronic disease risk prediction model training method and system. Background Art

[0002] With the aging of the global population and changing lifestyles, the incidence of chronic diseases continues to rise, posing significant challenges to public health systems and socioeconomics. Chronic diseases not only impact patients' quality of life but also increase the burden of healthcare. Effective chronic disease risk prediction can help healthcare institutions take preventive measures, reduce the probability of disease occurrence, and thus improve patients' quality of life and management outcomes.

[0003] Currently, most chronic disease risk prediction models are based on traditional statistical methods or simpler machine learning algorithms. These methods often rely on data features from a single modality, which limits the model's accuracy and generalization capabilities. Chronic disease risk is influenced by multiple factors, including genetics, environment, and behavior. Therefore, models that rely solely on a single data source or feature type struggle to capture the complex mechanisms of disease pathogenesis. Summary of the Invention

[0004] The purpose of the present invention is to provide a chronic disease risk prediction model training method and system to address the deficiencies in the existing technology, integrate multimodal data, implement effective feature selection and weighting, and realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

[0005] One embodiment of the present application provides a method for training a chronic disease risk prediction model, the method comprising:

[0006] Collect patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets in multiple modalities;

[0007] Based on an ensemble learning algorithm, multi-level feature selection is performed on the preliminary data feature set to screen for key features whose impact on chronic disease risk meets the preset criteria;

[0008] Use the attention mechanism to weight each key feature and generate corresponding weighted features;

[0009] Based on the weighted features, feature subsets at different levels are generated, and the model to be trained is layered trained using each feature subset, so that the model to be trained gradually learns the impact of different features on chronic disease risk prediction, and finally obtains a trained chronic disease risk prediction model, wherein a genetic algorithm is used to optimize the hyperparameter settings of the model to be trained.

[0010] Optionally, the ensemble learning algorithm is used to perform multi-level feature selection on the preliminary data feature set to screen key features whose impact on chronic disease risk meets preset standards, including:

[0011] Using random forest and gradient boosting tree techniques, the importance score of each feature is extracted to form a list of feature importance scores;

[0012] The preliminary data features are divided into three levels: high, medium, and low according to their importance scores. High-level preliminary data features are directly used as key features whose impact on chronic disease risk reaches the preset standard.

[0013] For the preliminary data features at the middle and low levels, recursive feature elimination technology is used to screen key features whose impact on the risk of chronic diseases meets the preset standards.

[0014] Optionally, the attention mechanism is used to weight each key feature to generate corresponding weighted features, including:

[0015] Using the attention mechanism, we determine the attention weight vector corresponding to each key feature vector, reflecting the importance of different modalities.

[0016] Calculate the product of each key feature vector and the corresponding attention weight vector respectively to obtain the corresponding weighted feature.

[0017] Optionally, generating feature subsets at different levels based on the weighted features, and performing hierarchical training on the to-be-trained model using each feature subset, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and ultimately obtains a trained chronic disease risk prediction model, including:

[0018] According to the level of key features contained in each weighted feature, each weighted feature is divided into feature subsets of different levels;

[0019] For each feature subset, a chronic disease risk prediction sub-model is trained according to the corresponding hierarchical order, so that the sub-model gradually adapts to and learns the impact of different features on chronic disease risk prediction. In each round of training, a feature subset is used as input and the sub-model is trained based on the feature subset.

[0020] The sub-models trained from each feature subset are integrated to obtain a comprehensive chronic disease risk prediction model.

[0021] Another embodiment of the present application provides a chronic disease risk prediction model training system, the system comprising:

[0022] A collection module, which is used to collect patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities;

[0023] The selection module is used to implement multi-level feature selection on the preliminary data feature set based on an ensemble learning algorithm to screen key features that have an impact on the risk of chronic diseases that meets the preset standards;

[0024] The weighting module is used to use the attention mechanism to weight each key feature and generate corresponding weighted features;

[0025] The training module is used to generate feature subsets of different levels based on the weighted features, and use each feature subset to perform hierarchical training on the model to be trained, so that the model to be trained can gradually learn the impact of different features on the prediction of chronic disease risks, and finally obtain a trained chronic disease risk prediction model, wherein the genetic algorithm is used to optimize the hyperparameter settings of the model to be trained.

[0026] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0027] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0028] Compared with the existing technology, the present invention provides a chronic disease risk prediction model training method, which collects patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities; based on the ensemble learning algorithm, multi-level feature selection is performed on the preliminary data feature set to screen key features whose impact on chronic disease risk reaches a preset standard; the attention mechanism is used to weight each key feature to generate corresponding weighted features; based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the to-be-trained model, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and finally obtains a trained chronic disease risk prediction model, wherein the genetic algorithm is used to optimize the hyperparameter settings of the to-be-trained model, so that multimodal data can be integrated, and effective feature selection and weighting are implemented to realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A hardware structure block diagram of a computer terminal for a chronic disease risk prediction model training method provided by an embodiment of the present invention;

[0030] Figure 2 A flowchart of a chronic disease risk prediction model training method provided by an embodiment of the present invention;

[0031] Figure 3 A schematic diagram of the structure of a chronic disease risk prediction model training system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0033] The embodiment of the present invention first provides a chronic disease risk prediction model training method, which can be applied to electronic devices such as computer terminals, specifically ordinary computers.

[0034] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for a chronic disease risk prediction model training method provided by an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown) a processor 102 (the processor 102 may include but is not limited to a microprocessor MCU or a programmable logic device FPGA and other processing devices) and a memory 104 for storing data. Optionally, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. For example, the computer terminal may also include Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0035] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / modules corresponding to the chronic disease risk prediction model training method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0036] The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by a communications provider of a computer terminal. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0037] See also Figure 2 , an embodiment of the present invention provides a chronic disease risk prediction model training method, which may include the following steps:

[0038] S201, collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities;

[0039] The key to this step is the data collection process. The heterogeneity of the data means that it comes from different types of sources, such as electronic health records (EHRs), medical images, genomic data, laboratory test results, social media data, and patients' self-reported lifestyle and health behaviors.

[0040] The core role of this step is to create a comprehensive multimodal feature set that covers the multi-dimensional information required for chronic disease prediction. In the risk prediction of chronic diseases, a single data source is often insufficient to capture the complex causes and related factors of the disease. By integrating data from multiple modalities, the model can gain a more comprehensive understanding of the patient's health status, thereby improving the accuracy and reliability of predictions.

[0041] 1. Diversity and comprehensiveness: Different data sources provide information from different perspectives, making the analysis more comprehensive and helping to capture various potential risk factors associated with chronic diseases.

[0042] 2. Improve data quality: By cross-validating patient information from different data sources, the accuracy and reliability of the data can be improved and data noise can be reduced.

[0043] 3. Integration of time and space dimensions: Different data sources may differ in time and space. By integrating these data, a more dynamic and three-dimensional patient health portrait can be established.

[0044] 4. Personalized medicine: The integration of multimodal data can better support personalized medicine and develop more targeted intervention measures. One way to achieve this can include:

[0045] 1. Determine the data source

[0046] Data source selection: Identify multiple heterogeneous data sources related to chronic diseases, such as electronic health records (EHRs), morbidity information, laboratory test results, imaging data, personal lifestyle surveys, socioeconomic background, and environmental factors.

[0047] Data source type: Different data sources can be structured (e.g., physiological indicators recorded in database tables) or unstructured (e.g., free-text comments from doctors, imaging data, etc.). By combining multiple data sources, a more comprehensive understanding of a patient's health status can be achieved.

[0048] 2. Data Acquisition and Extraction

[0049] Data capture tools: Utilize automated tools such as APIs and data crawlers to obtain relevant data from public health databases, hospital management systems, and scientific research data sharing platforms. Intrusive questionnaires can also be used to obtain information about patients' lifestyle habits and psychological assessments.

[0050] - Field definition and standardization: Define the fields and data types that need to be extracted (such as age, gender, blood pressure, blood sugar, family medical history, etc.), and ensure that the data extracted from different data sources remains consistent in units and formats so that subsequent analysis will not be affected by different data formats.

[0051] 3. Data fusion and cleaning

[0052] Data matching and fusion: Data from different data sources are matched using unique patient identifiers (such as national ID numbers and internal hospital IDs) to form a fused patient dataset. This step is crucial because it eliminates redundant and conflicting data records.

[0053] Data cleaning: In the fused dataset, perform missing value imputation, outlier detection and processing, and data duplication checks to ensure data quality. Missing values ​​can be handled using mean imputation, interpolation, or model-based imputation strategies.

[0054] 4. Preliminary feature extraction

[0055] - Feature generation: Extract preliminary features based on cleaned multimodal data. For example:

[0056] Physiological characteristics: Extract biomarkers of common diseases (such as blood sugar levels, cholesterol content, etc.) from EHR data.

[0057] - Lifestyle characteristics: Extract relevant data such as eating habits and exercise frequency from the questionnaire.

[0058] - Environmental characteristics: Use geographic information systems (GIS) to analyze the air quality, noise pollution, etc. of the patient's living environment.

[0059] Feature vectorization: Extracted features are vectorized to convert non-numeric data into numerical data for subsequent analysis. Categorical features can be represented as numerical features through methods such as one-hot encoding and word embedding, thus maintaining information integrity.

[0060] Through the above steps, we can effectively extract multimodal features related to chronic diseases from multiple heterogeneous data sources, laying a solid foundation for subsequent model training. This complex implementation method not only improves the performance of chronic disease risk prediction models, but also promotes the development of personalized medicine.

[0061] S202: Based on an ensemble learning algorithm, multi-level feature selection is performed on the preliminary data feature set to screen for key features whose impact on chronic disease risk meets the preset criteria;

[0062] In chronic disease risk prediction models, feature selection is a crucial step in data preprocessing. The goal of feature selection is to identify those features that significantly impact model results from a preliminary feature set, thereby reducing model complexity and improving prediction accuracy. At this stage, the application of ensemble learning algorithms can significantly improve the accuracy and stability of feature selection.

[0063] By combining the predictions of multiple base learners, ensemble learning algorithms can effectively overcome the shortcomings of a single algorithm in the feature selection process. By combining the advantages of multiple algorithms (such as random forests and gradient boosting trees), the impact of each feature on the risk of chronic diseases can be more comprehensively evaluated.

[0064] 1. Improve the accuracy of feature selection: Ensemble learning algorithms can integrate the results of multiple models, reduce the errors caused by the deviation of a single model, and thus improve the quality of feature selection.

[0065] 2. Reduce the risk of overfitting: By screening out more important features, the complexity of the model can be effectively reduced and the risk of overfitting can be reduced.

[0066] 3. Improve model interpretability: By focusing on key features, the interpretability of model results is enhanced, making it easier for doctors and medical practitioners to understand and apply them.

[0067] 4. Improve data utilization efficiency: By removing redundant features, the utilization efficiency of data during training can be effectively improved and the consumption of computing resources can be reduced.

[0068] Specifically, the importance score of each feature can be extracted using random forest and gradient boosting tree techniques to form a list of feature importance scores;

[0069] Random forest and gradient boosted tree (GBM) techniques are both very effective ensemble learning methods. They can construct a large number of decision trees, integrate prediction results, and calculate the importance score of each feature by counting the frequency of each feature in the tree branches. Such scores can effectively reflect the contribution of each feature to the prediction of chronic disease risk.

[0070] - Quantifying feature influence: Importance scores provide a quantitative basis for feature selection, helping to find the features that have the greatest impact on the target variable.

[0071] - Improve the targetedness of the model: By evaluating feature importance, we can focus more on features that have a greater impact on the risk of chronic diseases, which is helpful for subsequent analysis.

[0072] You can train a random forest and GBM model separately, using the same dataset and labels. For the random forest, the contribution of each feature to the split across multiple trees is calculated, summarizing the feature importance score. For the GBM, the gain (information gain) of each feature is calculated, forming a feature importance score. The feature importance scores from both models are summed to form a list of feature importance scores.

[0073] The preliminary data features are divided into three levels: high, medium, and low according to their importance scores. High-level preliminary data features are directly used as key features whose impact on chronic disease risk reaches the preset standard.

[0074] By dividing features according to their importance scores, we can clearly classify the influence of different features, which helps in subsequent feature selection and model training. The preset standard is based on the importance score of each feature. This process ensures that only the features with the greatest influence are selected for subsequent steps.

[0075] - Hierarchical feature management: High-level features can be immediately applied to subsequent models, while mid- and low-level features can be further screened.

[0076] - Improve training efficiency: Focus on important features and accelerate the model training process.

[0077] Based on the feature importance score list, you can set thresholds for high, medium, and low levels. By traversing the feature importance scores, you can categorize features into high, medium, and low levels based on the set thresholds. High-level features are directly used as key features for subsequent processing steps.

[0078] For the preliminary data features at the middle and low levels, recursive feature elimination technology is used to screen key features whose impact on the risk of chronic diseases meets the preset standards.

[0079] Recursive Feature Elimination (RFE) is a feature selection technique that repeatedly builds a model and removes the least important features until the required feature set remains. For mid-level and low-level features, RFE can help further screen out those that have a significant impact on chronic disease risk.

[0080] - Further optimize the feature set: By recursively filtering and removing redundant features, the dimension of the feature set can be reduced and the model performance can be optimized.

[0081] - Enhance model performance: Improve the robustness and accuracy of the model in predicting chronic disease risks.

[0082] When performing multi-level feature selection, the preset criteria can be designed to be hierarchical. For example:

[0083] - High-level features: Filter directly based on importance scores and select features with scores higher than the set threshold.

[0084] Mid-level and low-level features: Further screening using methods such as recursive feature elimination (RFE) may set stricter criteria to ensure that only those features that still perform well after multiple rounds of screening are retained. This process ensures that the prediction of chronic disease risk relies not only on the initial feature importance assessment, but also further verifies the validity of the features through subsequent recursive elimination methods.

[0085] In practical applications, preset criteria can also be manually set based on commonly known field experience. For example, if certain known biomarkers or clinical features are considered to be strongly associated with chronic diseases, these features can be prioritized in the selection process rather than relying solely on the calculated importance score.

[0086] Through the above steps, important features that affect the risk of chronic diseases can be effectively screened from multiple heterogeneous data sources, laying a solid foundation for the subsequent establishment of risk prediction models.

[0087] S203, using the attention mechanism to weight each key feature and generate corresponding weighted features;

[0088] In chronic disease risk prediction models, feature selection and weighting are crucial. Using the attention mechanism can effectively assign weights to different features to highlight the most representative features in predicting chronic disease risk. At this stage, the model will assign appropriate weights to each key feature based on its importance in overall risk prediction. This approach allows the model to pay more attention to certain features while reducing its focus on others during the learning process. Significance:

[0089] 1. Enhance the model's focusing ability: Through weighting, the model can better identify and utilize the most relevant features for chronic disease risk prediction, thereby improving prediction accuracy.

[0090] 2. Effectively handle multimodal data: In a multimodal feature set, the importance of different modalities may be inconsistent. The attention mechanism can assign appropriate weights to each modality, making the model more adaptable.

[0091] 3. Promote dynamic adjustment of features: During the training process, the model can dynamically adjust its focus based on the importance of features to adapt to data changes.

[0092] Specifically, the attention mechanism can be used to determine the attention weight vector corresponding to each key feature vector, which reflects the importance of different modalities;

[0093] The model calculates an attention weight vector based on each key feature vector. These weights reflect the relative importance of different modalities in the prediction task, helping the model focus on more influential features. These weights are typically obtained through training and adjusted using a specific loss function.

[0094] - Clarify the contribution of different modal features: Help understand the role of each modal feature in the prediction model.

[0095] - Dynamically optimize feature weights: Dynamically adjust feature weights based on the actual training results to enhance the model's adaptability.

[0096] Each key feature vector can be input into the model of the attention mechanism, and an attention weight vector is generated by applying an activation function (such as a softmax function) to each feature vector, in which each element represents the importance of the corresponding feature.

[0097] Calculate the product of each key feature vector and the corresponding attention weight vector to obtain the corresponding weighted feature.

[0098] In this step, the model multiplies each key feature vector by its corresponding attention weight element by element, and finally obtains the weighted features. These weighted features reflect the importance of each feature in the process of chronic disease risk prediction and can be used for subsequent model training.

[0099] - Highlight important features: Weighted features can highlight those features that contribute more to the prediction results, helping the model learn more effectively.

[0100] - Optimize model performance: By applying weights to features, the model can converge faster and improve training results.

[0101] The key feature vectors and corresponding attention weight vectors generated in the previous step can be collected. For each key feature vector, the corresponding weight vector is multiplied element-wise to obtain the weighted features. All weighted features are aggregated to form a new feature set for subsequent model training.

[0102] Using an attention mechanism to weight key features is an effective approach in training chronic disease risk prediction models. By calculating feature importance scores and combining them with dynamic weight adjustments, the model can focus on the most influential features, thereby improving overall predictive performance. This process not only improves model interpretability but also makes the model more adaptable, providing a more reliable basis for disease risk prediction.

[0103] S204, generating feature subsets of different levels based on the weighted features, and performing hierarchical training on the model to be trained using each feature subset, so that the model to be trained gradually learns the impact of different features on chronic disease risk prediction, and finally obtains a trained chronic disease risk prediction model, wherein the hyperparameter settings of the model to be trained are optimized using a genetic algorithm.

[0104] The core task of this stage is to organize the weighted features into layers. Weighted features are classified according to their importance and impact to form multiple feature subsets. During the layered training process, each feature subset is sequentially input into the model to be trained, allowing the model to gradually learn and grasp the predictive effects of different features on chronic disease risk. At the same time, the genetic algorithm is used to adaptively optimize the model's hyperparameter settings to ensure optimal model performance. Significance:

[0105] 1. Gradual Learning Mechanism: Through layered training, the model focuses on features at a specific level at a time, gradually improving its understanding and learning of complex relationships between features. This approach helps reduce the burden of model training and makes learning more efficient.

[0106] 2. Improve model accuracy: The layered training process can break the limitations of information mixing, allowing the model to fully absorb the input feature information in each round of training, thereby improving prediction accuracy.

[0107] 3. Optimizing model performance: The application of genetic algorithms ensures the optimization of model hyperparameters, reduces the complexity of manual parameter adjustment, and explores better configurations through evolutionary mechanisms, thereby further improving model performance.

[0108] Specifically, each weighted feature can be divided into feature subsets of different levels according to the level of key features contained in each weighted feature;

[0109] This step categorizes weighted features according to their importance and influence in chronic disease risk prediction. The hierarchical structure of feature subsets is set based on the weights of the features and is usually divided into high-level, medium-level, and low-level. This stratification helps the model focus on learning key features more effectively.

[0110] Optimizing feature learning: By dividing features into different levels, the model can focus on high-level features with greater impact, gradually introduce mid-level and low-level features, and enhance the effectiveness of learning.

[0111] - Improved training efficiency: Through layered training, the learning of features at each level no longer interferes with each other, making the training process more efficient and helping the model converge faster.

[0112] For each feature subset, a chronic disease risk prediction sub-model is trained according to the corresponding hierarchical order, so that the sub-model gradually adapts to and learns the impact of different features on chronic disease risk prediction. In each round of training, a feature subset is used as input and the sub-model is trained based on the feature subset.

[0113] In this step, for each feature subset, the sub-model is trained in a predetermined hierarchical order. Each training uses only one feature subset to ensure that the sub-model can learn the influence of each feature set without being interfered with by other features.

[0114] - Improve model accuracy: Gradual training allows the model to gain a deeper understanding of each feature subset, thereby improving the accuracy of the final prediction.

[0115] - Effective knowledge accumulation: Through step-by-step learning, the model can form a more systematic and comprehensive knowledge base, enhancing the understanding of chronic disease risks.

[0116] The sub-models trained from each feature subset are integrated to obtain a comprehensive chronic disease risk prediction model.

[0117] After all feature subsets are trained, the sub-models obtained are integrated to form a comprehensive prediction model. This comprehensive model can combine the advantages of each sub-model to improve the prediction accuracy of chronic diseases.

[0118] - Model ensemble effect: Ensembling different sub-models can effectively reduce the risk of overfitting and improve the stability and reliability of the model.

[0119] - Enhanced predictive capabilities: Through multi-model fusion, the learning results of various feature subsets can be integrated to ultimately form a powerful predictive tool.

[0120] An appropriate integration strategy can be determined, such as using simple weighted averaging, voting mechanisms, or stacking, to integrate the trained sub-models strategically to form the final chronic disease risk prediction model. This can combine the advantages of each sub-model and improve the accuracy and robustness of the overall prediction.

[0121] Additionally, during the training process, optimization techniques such as genetic algorithms can be used to optimize the ensemble model and its associated hyperparameter settings to achieve optimal model performance. Genetic algorithms generate fitness scores and select the optimal hyperparameter combination to further improve the model's predictive performance.

[0122] Specifically, multiple candidate hyperparameter combinations can be generated through mechanisms such as selection, crossover, and mutation. Each candidate solution is applied to the model training process, and its performance is evaluated (e.g., through cross-validation accuracy or loss function value). A fitness score is calculated based on the model's performance on the validation set, and the best-performing hyperparameter combination is selected for the next round of iterations until a predetermined number of generations is reached or the fitness improvement falls below a certain threshold.

[0123] Through the above steps, the chronic disease risk prediction model not only effectively learns practical features but also significantly improves its prediction accuracy and stability through layer-by-layer learning and model integration. Each step is designed to enable the model to more efficiently extract and utilize information when processing complex and heterogeneous data, providing strong support for chronic disease risk prediction.

[0124] It can be seen that patient data related to chronic diseases are collected from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities; based on the ensemble learning algorithm, multi-level feature selection is implemented on the preliminary data feature set to screen key features whose impact on chronic disease risks reaches the preset standards; the attention mechanism is used to weight each key feature to generate corresponding weighted features; based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the to-be-trained model, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and finally a trained chronic disease risk prediction model is obtained, wherein the genetic algorithm is used to optimize the hyperparameter settings of the to-be-trained model, so that multimodal data can be integrated, and effective feature selection and weighting are implemented to realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

[0125] Another embodiment of the present invention provides a chronic disease risk prediction model training system, see Figure 3 , the system may include:

[0126] A collection module 301 is used to collect patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities;

[0127] The selection module 302 is used to perform multi-level feature selection on the preliminary data feature set based on an ensemble learning algorithm to screen key features whose impact on chronic disease risk meets a preset standard;

[0128] The weighting module 303 is used to weight each key feature using the attention mechanism to generate corresponding weighted features;

[0129] The training module 304 is used to generate feature subsets of different levels based on the weighted features, and use each feature subset to perform hierarchical training on the model to be trained, so that the model to be trained can gradually learn the impact of different features on chronic disease risk prediction, and finally obtain a trained chronic disease risk prediction model, wherein the genetic algorithm is used to optimize the hyperparameter settings of the model to be trained.

[0130] It can be seen that patient data related to chronic diseases are collected from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities; based on the ensemble learning algorithm, multi-level feature selection is implemented on the preliminary data feature set to screen key features whose impact on chronic disease risks reaches the preset standards; the attention mechanism is used to weight each key feature to generate corresponding weighted features; based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the to-be-trained model, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and finally a trained chronic disease risk prediction model is obtained, wherein the genetic algorithm is used to optimize the hyperparameter settings of the to-be-trained model, so that multimodal data can be integrated, and effective feature selection and weighting are implemented to realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

[0131] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0132] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps:

[0133] S201, collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities;

[0134] S202: Based on an ensemble learning algorithm, multi-level feature selection is performed on the preliminary data feature set to screen for key features whose impact on chronic disease risk meets the preset criteria;

[0135] S203, using the attention mechanism to weight each key feature and generate corresponding weighted features;

[0136] S204, generating feature subsets of different levels based on the weighted features, and performing hierarchical training on the model to be trained using each feature subset, so that the model to be trained gradually learns the impact of different features on chronic disease risk prediction, and finally obtains a trained chronic disease risk prediction model, wherein the hyperparameter settings of the model to be trained are optimized using a genetic algorithm.

[0137] It can be seen that patient data related to chronic diseases are collected from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities; based on the ensemble learning algorithm, multi-level feature selection is implemented on the preliminary data feature set to screen key features whose impact on chronic disease risks reaches the preset standards; the attention mechanism is used to weight each key feature to generate corresponding weighted features; based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the to-be-trained model, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and finally a trained chronic disease risk prediction model is obtained, wherein the genetic algorithm is used to optimize the hyperparameter settings of the to-be-trained model, so that multimodal data can be integrated, and effective feature selection and weighting are implemented to realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

[0138] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0139] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0140] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0141] S201, collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities;

[0142] S202: Based on an ensemble learning algorithm, multi-level feature selection is performed on the preliminary data feature set to screen for key features whose impact on chronic disease risk meets the preset criteria;

[0143] S203, using the attention mechanism to weight each key feature and generate corresponding weighted features;

[0144] S204, generating feature subsets of different levels based on the weighted features, and performing hierarchical training on the model to be trained using each feature subset, so that the model to be trained gradually learns the impact of different features on chronic disease risk prediction, and finally obtains a trained chronic disease risk prediction model, wherein the hyperparameter settings of the model to be trained are optimized using a genetic algorithm.

[0145] It can be seen that patient data related to chronic diseases are collected from multiple heterogeneous data sources to generate preliminary data feature sets of multiple modalities; based on the ensemble learning algorithm, multi-level feature selection is implemented on the preliminary data feature set to screen key features whose impact on chronic disease risks reaches the preset standards; the attention mechanism is used to weight each key feature to generate corresponding weighted features; based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the to-be-trained model, so that the to-be-trained model gradually learns the impact of different features on chronic disease risk prediction, and finally a trained chronic disease risk prediction model is obtained, wherein the genetic algorithm is used to optimize the hyperparameter settings of the to-be-trained model, so that multimodal data can be integrated, and effective feature selection and weighting are implemented to realize hierarchical training of the model, thereby improving the accuracy and practicality of the chronic disease risk prediction model.

[0146] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A chronic disease risk prediction model training method, characterized in that: The method comprises: Collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets in multiple modalities; wherein the heterogeneous data sources are structured or unstructured data, and the heterogeneous data sources include: electronic health records, morbidity information, medical images, genomic data, laboratory test results, personal lifestyle surveys, socioeconomic background and environmental factors, social media data, and patient self-reported lifestyle and health behaviors; Based on an ensemble learning algorithm, multi-level feature selection was performed on the preliminary data feature set to screen for key features whose impact on chronic disease risk met the preset criteria. Random forest and gradient boosting tree techniques were used to extract the importance score of each feature and form a list of feature importance scores. The preliminary data features were divided into three levels: high, medium, and low based on their importance scores. High-level preliminary data features were directly used as key features whose impact on chronic disease risk met the preset criteria. Recursive feature elimination was used on the preliminary data features at the medium and low levels to screen for key features whose impact on chronic disease risk met the preset criteria. The attention mechanism is used to weight each key feature to generate corresponding weighted features. The attention mechanism is used to determine the attention weight vector corresponding to each key feature vector, which reflects the importance of different modalities. The product of each key feature vector and the corresponding attention weight vector is calculated to obtain the corresponding weighted feature. Based on the weighted features, feature subsets of different levels are generated, and each feature subset is used to perform hierarchical training on the model to be trained, so that the model to be trained gradually learns the impact of different features on chronic disease risk prediction, and finally a trained chronic disease risk prediction model is obtained, wherein a genetic algorithm is used to optimize the hyperparameter settings of the model to be trained; wherein, according to the level of key features contained in each weighted feature, each weighted feature is divided into feature subsets of different levels; for each feature subset, a chronic disease risk prediction sub-model is trained according to the hierarchical order, so that the sub-model gradually adapts to and learns the impact of different features on chronic disease risk prediction, in each round of training, a feature subset is used as input, and the sub-model is trained based on the feature subset; the sub-models trained by each feature subset are integrated to obtain a comprehensive chronic disease risk prediction model.

2. A chronic disease risk prediction model training system, characterized in that: The system comprises: A collection module for collecting patient data related to chronic diseases from multiple heterogeneous data sources to generate preliminary data feature sets in multiple modalities; wherein the heterogeneous data sources are structured data or unstructured data, and the heterogeneous data sources include: electronic health records, morbidity information, medical images, genomic data, laboratory test results, personal lifestyle surveys, socioeconomic background and environmental factors, social media data, and patient self-reported lifestyle and health behaviors; A selection module is used to perform multi-level feature selection on the preliminary data feature set based on an ensemble learning algorithm to screen key features whose impact on chronic disease risk meets the preset standard. The module uses random forest and gradient boosting tree techniques to extract the importance score of each feature and form a feature importance score list. The preliminary data features are divided into three levels: high, medium, and low according to the importance score. High-level preliminary data features are directly used as key features whose impact on chronic disease risk meets the preset standard. For medium-level and low-level preliminary data features, recursive feature elimination technology is used to screen key features whose impact on chronic disease risk meets the preset standard. A weighting module is used to weight each key feature using the attention mechanism to generate corresponding weighted features. The attention mechanism is used to determine the attention weight vector corresponding to each key feature vector, which reflects the importance of different modalities. The product of each key feature vector and the corresponding attention weight vector is calculated to obtain the corresponding weighted feature. A training module is used to generate feature subsets of different levels based on the weighted features, and use each feature subset to perform hierarchical training on the model to be trained, so that the model to be trained can gradually learn the impact of different features on chronic disease risk prediction, and finally obtain a trained chronic disease risk prediction model, wherein a genetic algorithm is used to optimize the hyperparameter settings of the model to be trained; wherein, according to the level of key features contained in each weighted feature, each weighted feature is divided into feature subsets of different levels; for each feature subset, a chronic disease risk prediction sub-model is trained according to the hierarchical order, so that the sub-model can gradually adapt to and learn the impact of different features on chronic disease risk prediction, and in each round of training, a feature subset is used as input, and the sub-model is trained based on the feature subset; the sub-models obtained by training each feature subset are integrated to obtain a comprehensive chronic disease risk prediction model.

3. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to claim 1 when executed.

4. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to claim 1 .

Images