Blood glucose state prediction method, system and terminal based on integrated learning model

Abstract

The invention discloses a blood glucose state prediction method and system based on an integrated learning model and a terminal, and the method comprises the steps: constructing a data set, preprocessing the data set, obtaining a preprocessed data set, constructing the integrated learning model through an integration strategy based on the preprocessed data set, and training the integrated learning model. Obtaining a target blood glucose state prediction model; collecting target basic information and target conventional blood examination information of a to-be-predicted user, inputting the target basic information and the target conventional blood examination information into the target blood glucose state prediction model for prediction, and outputting a blood glucose state prediction result based on the integrated learning model; generating a personalized prompt according to the blood glucose state prediction result based on the integrated learning model, and sending the blood glucose state prediction result and the personalized prompt to the user; the invention innovatively provides an integrated learning model which combines various data processing methods with an integrated strategy, and improves the accuracy of blood glucose state prediction.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a blood sugar state prediction method, system, terminal and computer-readable storage medium based on an integrated learning model. Background Art

[0002] Blood sugar, also often referred to as blood glucose, refers to the content of glucose in the blood. It is an important indicator for assessing the normality of glucose metabolism and can be used to screen for a variety of abnormalities in glucose metabolism.

[0003] With the development of various neural network technologies, models for predicting blood sugar status have been proposed. However, the data sets of these models mostly contain only blood sugar-related indicators, and do not use indicators unrelated to blood sugar to complete the prediction of human blood sugar status; secondly, the design of existing blood sugar prediction models lacks more effective integration strategies and methods to improve model accuracy, resulting in inaccurate prediction results for blood sugar status.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] The main purpose of the present invention is to provide a blood glucose status prediction method, system, terminal and computer-readable storage medium based on an integrated learning model, aiming to solve the problem that the indicators used to predict blood glucose data in the prior art are not sufficient, and the design of the blood glucose prediction model lacks a more effective integration strategy, resulting in the current prediction results of blood glucose status being inaccurate.

[0006] To achieve the above object, the present invention provides a method for predicting blood sugar status based on an ensemble learning model, the method comprising the following steps: Constructing a data set, preprocessing the data set to obtain a preprocessed data set, constructing an integrated learning model based on the preprocessed data set through a preset integration strategy, and training the integrated learning model to obtain a target blood glucose state prediction model; Collecting target basic information and target routine blood test information of the user to be predicted, inputting the target basic information and the target routine blood test information into the target blood glucose state prediction model through encryption technology for prediction, and outputting the blood glucose state prediction result based on the integrated learning model; A personalized reminder is generated according to the blood sugar status prediction result based on the ensemble learning model, and the blood sugar status prediction result based on the ensemble learning model and the personalized reminder are sent to the user to be predicted, wherein the personalized reminder includes screening suggestions and lifestyle guidance.

[0007] Optionally, in the blood glucose status prediction method based on an integrated learning model, the step of constructing a data set specifically includes: Collecting historical basic information and historical routine blood test information from the data system as original features, obtaining blood sugar states corresponding to the historical basic information and the historical routine blood test information as labels, and constructing an original data set based on the original features and the labels; De-identifying the data in the original data set, removing the data that reflects personal identity information, and performing access control and encryption on the removed data to obtain the data set; The historical basic information includes age information, gender information, height information and weight information, and the historical routine blood test information includes complete blood cell count information, liver function information, kidney function information and blood lipid information.

[0008] Optionally, the blood sugar state prediction method based on the ensemble learning model, wherein the preprocessing of the data set to obtain the preprocessed data set specifically includes: Using reference values ​​of indicators corresponding to gender and age to perform feature normalization processing on the data set, unifying the dimension and range of the features, and obtaining a first intermediate data set; Using the isolation forest algorithm to calculate the average path length of each data point in the first intermediate data set, calculating the anomaly score of each data point according to the average path length, taking the anomaly score higher than a preset threshold as an outlier, and removing the data points corresponding to the outlier to obtain a second intermediate data set; Generate new minority class samples using the SMOTE algorithm, synthesize the minority class samples with the second intermediate data set to obtain synthesized samples, clean the synthesized samples using the Tomek link technology, remove overlapping sample data and noise sample data on the boundary, and generate a third intermediate data set; The third intermediate data set is subjected to collinearity detection using a correlation coefficient matrix detection method and a variance inflation factor detection method, and the detected collinear variables are removed from the third intermediate data set to obtain a preprocessed data set.

[0009] Optionally, in the blood glucose status prediction method based on an integrated learning model, the preset integrated strategy includes a multi-objective iterative model selection method and a stacking method including a hidden layer neural network; The step of constructing an integrated learning model based on the preprocessed data set through a preset integration strategy specifically includes: Constructing a candidate model set, and generating an integrated model through the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set; The integrated model is stacked by a stacking method including a hidden layer neural network to generate an integrated learning model of a stacked meta-learner.

[0010] Optionally, the blood glucose state prediction method based on the ensemble learning model, wherein the step of constructing a candidate model set, generating an ensemble model by the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set, specifically comprises: Acquire multiple initial candidate models, select hyperparameters for each of the initial candidate models using a Bayesian optimization algorithm to obtain candidate models, and construct a model candidate set including the multiple candidate models; Using the preprocessed data set to cross-validate each candidate model in the model candidate set, generating a prediction result for each candidate model, calculating the individual accuracy of each candidate model based on a plurality of the prediction results, and adding the candidate model with the highest individual accuracy from the model candidate set to the new model subset; Calculating the diversity of all remaining candidate models in the model candidate set relative to the model subset, normalizing the diversity and individual accuracy corresponding to each remaining candidate model, and obtaining the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model; Calculating the cost functions of all remaining candidate models according to the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model, and adding the remaining candidate model with the minimum cost function from the model candidate set to the model subset; Using the model subset to construct an initial integrated model, and using the preprocessed data set to cross-validate the integrated model, if the performance improvement of the initial integrated model does not exceed a preset threshold, the initial integrated model is used as the integrated model; If the performance improvement of the initial integrated model exceeds a preset threshold, continue to select the remaining candidate models with the minimum cost function from the model candidate set to add to the model subset until the performance of the initial integrated model no longer improves or the improvement is less than the preset improvement value, remove the remaining candidate model last added to the initial integrated model, and obtain the integrated model.

[0011] Optionally, the blood sugar state prediction method based on the ensemble learning model, wherein the ensemble model is stacked by a stacking method including a hidden layer neural network to generate an ensemble learning model of a stacked meta-learner, specifically comprises: The preprocessed data set is divided into a training set and a test set according to a preset ratio, and each base learner in the integrated model is trained and predicted by a K-fold cross-validation algorithm according to the training set to obtain a first prediction result of each base learner on the training set, a primary feature matrix is ​​generated according to the first prediction result, and a second prediction result is generated using the test set; Wherein, each base learner in the integrated model corresponds to a model in the integrated model; Using the primary feature matrix as a training feature of a meta-learner, using the second prediction result as a test feature of the meta-learner, and using the label of the training set as a training label and a test label of the meta-learner to construct a meta-learner data set; Constructing an ensemble learning model, setting an input layer of the ensemble learning model, adding a preset number of meta-learners after the input layer, and setting an output layer of the ensemble learning model according to the task type; The number of nodes in the input layer is equal to the number of base learners, and each node corresponds to the first prediction result of a base learner.

[0012] Optionally, the blood glucose state prediction method based on the ensemble learning model, wherein the training of the ensemble learning model to obtain a target blood glucose state prediction model specifically includes: The ensemble learning model is trained and tested using the meta-learner dataset, and the prediction of the ensemble learning model is decomposed into multiple interpretable parts using Shapley additive explanation, each of which represents the contribution of a feature in the current sample to achieve an evaluation of the importance of the feature; The performance index of the integrated learning model is calculated, and if the performance index meets the preset conditions, a target blood glucose state prediction model is obtained.

[0013] In addition, to achieve the above-mentioned purpose, the present invention also provides a blood sugar state prediction system based on an integrated learning model, wherein the blood sugar state prediction system based on an integrated learning model comprises: A target model training module is used to construct a data set, preprocess the data set to obtain a preprocessed data set, construct an integrated learning model based on the preprocessed data set through a preset integration strategy, and train the integrated learning model to obtain a target blood sugar state prediction model; A blood sugar status prediction module, which is used to collect the target basic information and target routine blood test information of the user to be predicted, input the target basic information and the target routine blood test information into the target blood sugar status prediction model through encryption technology for prediction, and output the blood sugar status prediction result based on the integrated learning model; A reminder and notification module is used to generate personalized reminders according to the blood sugar status prediction results based on the integrated learning model, and send the blood sugar status prediction results based on the integrated learning model and the personalized reminders to the user to be predicted, wherein the personalized reminders include screening suggestions and lifestyle guidance.

[0014] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a blood glucose status prediction program based on an integrated learning model stored in the memory and executable on the processor, wherein the blood glucose status prediction program based on the integrated learning model, when executed by the processor, implements the steps of the blood glucose status prediction method based on the integrated learning model as described above.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a blood glucose status prediction program based on an integrated learning model, and when the blood glucose status prediction program based on an integrated learning model is executed by a processor, the steps of the blood glucose status prediction method based on an integrated learning model as described above are implemented.

[0016] In the present invention, a data set is constructed, and the data set is preprocessed to obtain a preprocessed data set. Based on the preprocessed data set, an integrated learning model is constructed through an integrated strategy, and the integrated learning model is trained to obtain a target blood glucose state prediction model; target basic information and target routine blood test information of the user to be predicted are collected, and the target basic information and target routine blood glucose state prediction model are input into the target blood glucose state prediction model for prediction, and the blood glucose state prediction result based on the integrated learning model is output; personalized reminders are generated according to the blood glucose state prediction results based on the integrated learning model, and the blood glucose state prediction results and personalized reminders are sent to the user; the present invention innovatively proposes an integrated learning model that combines multiple data processing methods with an integrated strategy, thereby improving the accuracy of blood glucose state prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a preferred embodiment of the blood sugar state prediction method based on the integrated learning model of the present invention; Figure 2 It is a flowchart of preprocessing the data set in the method for predicting blood sugar status based on an integrated learning model of the present invention; Figure 3 It is a flow chart of generating an integrated model by the multi-objective iterative model selection method in the blood sugar state prediction method based on the integrated learning model of the present invention; Figure 4 It is a structural diagram of a preferred embodiment of a blood sugar state prediction system based on an integrated learning model of the present invention; Figure 5It is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0018] The present application provides a method, system and terminal for predicting blood glucose status based on an integrated learning model. In order to make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0020] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0021] The blood sugar state prediction method based on the ensemble learning model described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the blood sugar state prediction method based on the integrated learning model includes the following steps: Step S10, constructing a data set, preprocessing the data set to obtain a preprocessed data set, constructing an integrated learning model based on the preprocessed data set through a preset integration strategy, training the integrated learning model, and obtaining a target blood glucose state prediction model.

[0022] The constructing of the data set specifically includes: Collecting historical basic information and historical routine blood test information from the data system as original features, obtaining blood sugar states corresponding to the historical basic information and the historical routine blood test information as labels, and constructing an original data set based on the original features and the labels; De-identifying the data in the original data set, removing the data that reflects personal identity information, and performing access control and encryption on the removed data to obtain the data set; The historical basic information includes age information, gender information, height information and weight information, and the historical routine blood test information includes complete blood cell count information, liver function information, kidney function information and blood lipid information.

[0023] In this embodiment, the original data set construction method is suitable for model developers who do not have ready-made data. They need to build a data set through a data system (such as an electronic health record of a medical institution). The first step in building a data set is to select original features and labels. Features are input variables used to describe data, also commonly referred to as independent variables or inputs. Each feature represents a certain attribute or dimension of the data and is the basis for the model to learn and predict. Original features refer to unscreened features, and the focus is on obtaining as many types of features as possible. When acquiring original features, all or part of the routine blood test indicators can be selected according to the data recorded in the data system, including complete blood count, liver function, kidney function, blood lipids and other historical routine blood test information, as well as basic information: age, gender, height, and weight.

[0024] The label is the target variable that the model needs to predict, and is also usually called the dependent variable or output. This is the part that the model learns during training, and is used to tell the model what to predict. In this embodiment, the blood sugar state corresponding to the historical basic information and the historical routine blood test information (blood sugar state includes hyperglycemia state, normal blood sugar state and hypoglycemia state) is obtained as a label.

[0025] The privacy protection of data in the data system is still a top priority. First of all, the data must be de-identified, that is, the information that can directly or indirectly identify an individual, such as name, ID number, address, telephone number, etc., must be removed or obfuscated. By replacing personal identifiers with anonymous IDs, or obfuscating personal birth dates and addresses, it is ensured that the data cannot be associated with specific personal information. De-identification can effectively reduce the impact of data leakage on personal privacy. Next, access control must be carried out. Only authorized personnel can access the data, and access rights should be strictly graded according to responsibilities. By setting up multi-level permission control, it is ensured that only staff who need to operate the data can view and process relevant information. In addition, although the data in the data system is offline data and does not involve encryption issues for network transmission, the media for data storage still needs to be properly protected. Hard disk encryption or physical isolation can be used to ensure the security of data storage. During the operation process, copying and dissemination of unprocessed original data should be avoided as much as possible.

[0026] Furthermore, if Figure 2 As shown, the data set is preprocessed to obtain a preprocessed data set, which specifically includes: S101: performing feature normalization processing on the data set using reference values ​​of indicators corresponding to gender and age, unifying the dimension and range of the features, and obtaining a first intermediate data set.

[0027] It is understandable that since different features may have different dimensions and ranges, this may affect the training effect of the model, so the data needs to be normalized. The present invention uses the reference values ​​of the indicators corresponding to gender and age to perform feature normalization: ; in, represents the feature value in the data set, Indicates the normalized value. Indicates that the measurement is gender specific. and age The reference range below is derived from the standardized reference range published for a certain blood index. represents the upper limit of the reference range, Indicates the lower bound of the reference range.

[0028] For example, if an adult male has a red blood cell count of 5.0 million / mm³ and the reference range is [4.30, 5.70], then according to the above formula: subtract the lower limit of the reference range, 4.30, from 5.0 to get 0.70, and then divide 0.70 by the difference in the reference range, 1.40 (i.e. 5.70 - 4.30), to get the normalized result of 0.5.

[0029] S102: Use the isolation forest algorithm to calculate the average path length of each data point in the first intermediate data set, calculate the anomaly score of each data point based on the average path length, take the anomaly score higher than a preset threshold as an outlier, remove the data points corresponding to the outlier, and obtain the second intermediate data set.

[0030] It is understandable that the processing of constant values ​​is also a key point in data preprocessing. Outliers are usually caused by measurement errors or input errors. Outlier detection uses the Isolation Forest (iForest) algorithm, which is based on the principle of quantifying the "abnormality" of each point by constructing multiple isolation trees (iTree) and using the split path length of these trees. If a data point is quickly isolated in the tree (shorter path), then this data point may be an outlier; if a data point requires more steps to be isolated (longer path), then this data point may be a normal point.

[0031] The algorithm flow of Isolation Forest is as follows: Input data set (Dataset The size is n ), the number of trees and the maximum number of samples per tree ( Generally less than , usually 256); then, construct an isolation forest: randomly select A subset of samples, construct Isolation tree, for each data point , calculate the path length in each tree ; Then calculate the average path length of the data point in all trees : .

[0032] Furthermore, the anomaly score of each data point is calculated based on the expected value of the path length : ; Then the data points are classified as normal points or abnormal points according to the anomaly score, : Indicates that the point is an outlier (the path length is short), : Indicates that the point is a normal point. : Indicates that the path length of the point is very long, and it is very likely to be a normal point; by setting a threshold, the data points can be classified, and 0.5 is usually used as the default threshold. Finally, the anomaly score of each data point is output. Among them, is an expected value used to normalize the path length.

[0033] S103: Use the SMOTE algorithm to generate new minority class samples, synthesize the minority class samples with the second intermediate data set to obtain synthesized samples, use Tomek link technology to clean the synthesized samples, remove overlapping sample data and noise sample data on the boundary, and generate a third intermediate data set.

[0034] There is a common imbalance in the collected data. Generally speaking, the ratio of positive and negative samples in imbalanced data is extremely different, and imbalanced data will affect the practical application value of the model, so the data needs to be balanced.

[0035] In this embodiment, the SMOTE Tomek algorithm is used, which is a technology that combines SMOTE (synthetic minority oversampling technology) and Tomek link (data cleaning method) to deal with the imbalance problem in the data set and clean up the noise samples on the classification boundary; the core idea of ​​SMOTE is to generate new minority class samples by interpolation instead of simply copying existing samples. In this way, the distribution of minority class samples can be expanded without causing overfitting.

[0036] The step of using the SMOTE algorithm to generate new minority class samples and synthesizing the minority class samples with the second intermediate data set to obtain synthesized samples includes: Finding minority class samples of Nearest neighbor samples: For sample , using some distance metric (such as Euclidean distance) to find of nearest neighbor minority class samples; Randomly select a nearest neighbor sample :from of Randomly select a sample from the nearest neighbor samples ; Interpolation generates new samples :pass and Interpolation between generates a new synthetic sample : ;in, is a random number in the range [0, 1], used to control the interpolation position. is the generated synthetic sample; Through the above interpolation formula, the synthetic samples generated by SMOTE lie in and , thereby expanding the distribution of the minority class.

[0037] Furthermore, Tomek link is a technique for cleaning the boundaries of a dataset by detecting the nearest neighbor relationship between samples of different categories, finding samples that are on the classification boundary, and removing them from the dataset. Specifically, when two samples and A Tomek link is formed when the following conditions are met: and belong to different categories, and and As the nearest neighbors of each other, these samples are usually located at the boundary of the category, which can easily introduce noise and confusion.

[0038] The method of using Tomek linking technology to clean up the synthetic samples, remove overlapping sample data and noise sample data on the boundary, and generate a third intermediate data set includes: Find Tomek links: For each pair of samples in the dataset ,calculate and If the sample and another sample Belong to different categories, and the two are each other's nearest neighbors, then the sample and Constructs a Tomek link.

[0039] Remove samples: For each Tomek link, you can choose to remove samples of the majority class or , to reduce category overlap and clean up noisy samples.

[0040] S104: Use a correlation coefficient matrix detection method and a variance inflation factor detection method to perform collinearity detection on the third intermediate data set, remove the detected collinear variables from the third intermediate data set, and obtain a preprocessed data set.

[0041] Collinearity (also called multicollinearity) refers to the situation where there is a high correlation between features. This high correlation will lead to the instability of subsequent model parameters, and the estimated value of the regression coefficient may become very large or very inaccurate, thereby affecting the explanatory power and predictive ability of the model. In order to detect and solve the collinearity problem, the present invention uses a correlation coefficient matrix detection method and a variance inflation factor detection method to perform collinearity detection on the third intermediate data set, and removes the detected collinear variables from the third intermediate data set to obtain a preprocessed data set.

[0042] It is understandable that the present invention solves the collinearity problem by deleting one of the independent variables that is highly correlated with other variables or removing the independent variables with higher VIF values. In addition, highly correlated variables can be combined into a new variable; high-dimensional data can be reduced in dimension by principal component analysis; regularized regression (such as ridge regression) can be used, etc., which are not limited here.

[0043] Furthermore, the preset integration strategy includes a multi-objective iterative model selection method and a stacking method including a hidden layer neural network.

[0044] The step of constructing an integrated learning model based on the preprocessed data set through a preset integration strategy specifically includes: Constructing a candidate model set, and generating an integrated model through the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set; The integrated model is stacked by a stacking method including a hidden layer neural network to generate an integrated learning model of a stacked meta-learner.

[0045] In this embodiment, the integration strategy includes two parts: one is model selection, which uses a multi-objective iterative model selection (MoItMS) method; the second is model fusion, which uses a stacking method involving hidden layer neural networks.

[0046] The multi-objective iterative model selection method is a model selection strategy for ensemble learning. It optimizes the individual accuracy and model diversity of the model and iteratively selects the best model combination to improve the overall performance of the ensemble model. The MoItMS method mainly optimizes individual accuracy and model diversity, where individual accuracy is used to measure the prediction performance of a single model, and model diversity is used to measure the differences between models to ensure the generalization ability of the ensemble model.

[0047] like Figure 3 As shown, the constructing of the candidate model set, generating an integrated model through the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set, specifically includes: Step S201: obtain multiple initial candidate models, select hyperparameters for each of the initial candidate models using a Bayesian optimization algorithm to obtain candidate models, and construct a model candidate set including the multiple candidate models.

[0048] The initial candidate models selected are the following machine learning models: Multilayer Perceptron (MLP), K-Nearest Neighbors (KNN), Decision Tree (DT), Support Vector Machine (SVM), Logistic Regression Model (LRM), Random Forest (RF), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM).

[0049] Furthermore, a Bayesian optimization algorithm is used to select hyperparameters for each of the initial candidate models. Bayesian optimization is a strategy for global optimization, particularly suitable for optimization problems of black box functions. An important application scenario of Bayesian optimization in machine learning is hyperparameter selection, which can effectively find the best hyperparameter combination of the model under limited computing resources. Unlike traditional grid search or random search, Bayesian optimization approximates the objective function by constructing a proxy model (usually a Gaussian process) and searches for the optimal solution in uncertainty. This method can find the globally optimal hyperparameters in fewer iterations by balancing exploration and exploitation.

[0050] Step S202: use the preprocessed data set to cross-validate each candidate model in the model candidate set, generate a prediction result for each candidate model, calculate the individual accuracy of each candidate model based on multiple prediction results, and add the candidate model with the highest individual accuracy from the model candidate set to the new model subset.

[0051] In this embodiment, first, the model candidate set is processed using the preprocessed data set. Each of the candidate models in is cross-validated to generate a prediction result for each of the candidate models, where Indicates characteristics, Indicates a label.

[0052] According to the multiple prediction results, the individual accuracy of each candidate model is calculated by the mean square error: ; in, Represents candidate models The individual accuracy of represents the number of samples, Represents candidate models For The predicted value of samples, Indicates The true labels of the samples.

[0053] Step S203, calculating the diversity of all remaining candidate models in the model candidate set relative to the model subset, normalizing the diversity and individual accuracy corresponding to each remaining candidate model, and obtaining the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model.

[0054] In order to make individual precision and diversity comparable, they need to be normalized: ; ; in, and are the maximum individual accuracy and maximum diversity among all remaining candidate models, and The normalized individual accuracy of each remaining candidate model is and diversity .

[0055] Step S204: Calculate the cost function of all remaining candidate models according to the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model, and add the remaining candidate model with the minimum cost function from the model candidate set to the model subset.

[0056] According to the normalized diversity corresponding to each remaining candidate model and normalized individual accuracy , calculate the cost function of all remaining candidate models : ; in, is a weight parameter used to balance the importance of individual accuracy and diversity.

[0057] Further, the remaining candidate model with the minimum cost function is selected from the remaining candidate models, and the remaining candidate model with the minimum cost function is added from the model candidate set to the model subset.

[0058] Step S205: construct an initial integrated model using the model subset, and cross-validate the integrated model using the preprocessed data set; if the performance improvement of the initial integrated model does not exceed a preset threshold, use the initial integrated model as the integrated model.

[0059] In this embodiment, the model subset is used Construct an initial ensemble model and evaluate its performance: Evaluate the performance of the initial ensemble model through cross-validation. If the performance improvement of the initial ensemble model does not exceed the preset threshold, it means that the remaining candidate models cannot bring significant performance improvement to the ensemble model. In this case, the initial ensemble model is directly used as the ensemble model.

[0060] Step S206: If the performance improvement of the initial integrated model exceeds a preset threshold, continue to select the remaining candidate models with the minimum cost function from the model candidate set to add to the model subset until the performance of the initial integrated model is no longer improved or the improvement is less than the preset improvement value, remove the remaining candidate model last added to the initial integrated model, and obtain the integrated model.

[0061] It can be understood that if the performance improvement of the initial integrated model exceeds a preset threshold, the process of steps S203-S204 above is iterated, and the remaining candidate models with the minimum cost function are continuously selected from the model candidate set to be added to the model subset, and the performance of the re-formed integrated model is evaluated until the performance improvement of the integrated model is insufficient or not improved, the last candidate model added is removed, and the iteration ends.

[0062] Furthermore, another part of the integration strategy: a stacking method containing hidden layer neural networks is an advanced integrated learning method that further improves the performance of the model by training a meta learner using the prediction results of multiple base learners as new features. Using a neural network containing hidden layers as a meta learner can take advantage of its powerful nonlinear modeling capabilities to better capture the complex relationship between the prediction results of the base learners and improve the generalization ability of the integrated model.

[0063] The stacking method including a hidden layer neural network is used to stack the integrated model to generate an integrated learning model of a stacked meta-learner, specifically including: The preprocessed data set is divided into a training set and a test set according to a preset ratio, and each base learner in the integrated model is trained and predicted by a K-fold cross-validation algorithm according to the training set to obtain a first prediction result of each base learner on the training set, a primary feature matrix is ​​generated according to the first prediction result, and a second prediction result is generated using the test set; Wherein, each base learner in the integrated model corresponds to a model in the integrated model; Using the primary feature matrix as a training feature of a meta-learner, using the second prediction result as a test feature of the meta-learner, and using the label of the training set as a training label and a test label of the meta-learner to construct a meta-learner data set; Constructing an ensemble learning model, setting an input layer of the ensemble learning model, adding a preset number of meta-learners after the input layer, and setting an output layer of the ensemble learning model according to the task type; The number of nodes in the input layer is equal to the number of base learners, and each node corresponds to the first prediction result of a base learner.

[0064] It can be understood that in the stacking method of the present invention, the structure of the integrated learning model includes a base learner (Level-0 model) and a meta learner (Level-1 model), wherein the base learner is a plurality of different types of machine learning models (such as decision trees, support vector machines, random forests, etc.), and the plurality of different types of machine learning models respectively learn and predict the original data. In the present invention, the base learner is used for the model subset obtained by the multi-objective iterative model selection method. The meta-learner is a neural network with hidden layers. The input is the prediction results of the base learner on the training set and the test set, and the output is the final prediction result.

[0065] The specific process of the K-fold cross-validation algorithm is to divide the training set into K subsets. In each iteration, the model is trained using K-1 subsets, and prediction is performed on the remaining subset. This is repeated K times to ensure that each subset is predicted once as a validation set. In this way, the prediction results of the base learner for each sample in the training set are obtained to form a "primary feature matrix".

[0066] It can be understood that the output layer of the integrated learning model is set according to the task type. For classification tasks, a Softmax or Sigmoid activation function is used; for regression tasks, a continuous value is output.

[0067] Furthermore, the ensemble learning model is trained to obtain a target blood glucose state prediction model. The trained neural network is used to predict the test features of the meta-learner to obtain the final prediction results, and then the performance indicators of the model are calculated, such as accuracy, AUC value, F1 score, etc. Finally, the performance of the stacked model is compared with the performance of each base learner to evaluate whether the stacked model has achieved performance improvement and verify the effectiveness of the stacking method using a hidden layer neural network.

[0068] The training of the integrated learning model to obtain a target blood glucose state prediction model specifically includes: The ensemble learning model is trained and tested using the meta-learner dataset, and the prediction of the ensemble learning model is decomposed into multiple interpretable parts using Shapley additive explanation, each of which represents the contribution of a feature in the current sample to achieve an evaluation of the importance of the feature; The performance index of the integrated learning model is calculated, and if the performance index meets the preset conditions, a target blood glucose state prediction model is obtained.

[0069] In this embodiment, the ensemble learning model is trained and tested using the meta-learner dataset, and the following methods are used to evaluate the effect of the model: precision, specificity, recall or sensitivity, accuracy, F1-measure, and area under the curve; and Shapley Additive exPlanations (SHAP) is used to evaluate the importance of features. SHAP is a tool for explaining machine learning models, which allocates the contribution of each feature to model prediction based on the Shapley value in game theory. The main goal of SHAP is to provide a local explanation for the prediction of each sample, so that the role played by each feature in the prediction of this sample can be known.

[0070] Step S20: collect the target basic information and target routine blood test information of the user to be predicted, input the target basic information and the target routine blood test information into the target blood glucose status prediction model through encryption technology for prediction, and output the blood glucose status prediction result based on the integrated learning model.

[0071] Specifically, the target basic information and target routine blood test information of the predicted user are collected, and all transmitted data (target basic information and target routine blood test information) are encrypted using the Transport Layer Security (TLS) or Secure Sockets Layer (SSL) to ensure that the data will not be stolen or tampered with by a third party during transmission.

[0072] Furthermore, before data transmission, symmetric encryption algorithms (Advanced Encryption Standard, AES-256) can be used to encrypt data. Even if the data is intercepted during transmission, attackers cannot decrypt the encrypted data. For particularly sensitive information (such as identity data), asymmetric encryption algorithms (such as RSA encryption or elliptic curve encryption) are used for additional encryption.

[0073] Furthermore, the target blood sugar state prediction model is called, the encrypted and transmitted target basic information and target routine blood test information are input into the target blood sugar state prediction model for online reasoning, and the blood sugar state prediction result is returned immediately.

[0074] Step S30, generating a personalized reminder according to the blood sugar status prediction result based on the ensemble learning model, and sending the blood sugar status prediction result based on the ensemble learning model and the personalized reminder to the user to be predicted, wherein the personalized reminder includes screening suggestions and lifestyle guidance.

[0075] In this embodiment, in order to ensure timely processing of messages, the generated blood sugar status prediction results will be immediately conveyed to the user to be predicted and the medical staff, and personalized reminders will be generated based on the blood sugar status prediction results. The personalized reminders are used to provide screening recommendations and lifestyle guidance. The blood sugar status prediction results and the personalized reminders can be sent via App, SMS, email, etc.

[0076] It can be seen that the data set construction for adding opportunistic screening proposed in the present invention selects routine blood test data and basic data of users as the input of the model, and the data is more common, easy to obtain and promote; at the same time, an integrated learning model combining multiple data processing methods with integrated strategies is innovatively proposed, including: gender and age reference values ​​for standardization, isolation forest algorithm for outlier detection, SMOTE Tomek for unbalanced feature data processing, collinearity removal method, multi-objective iterative model selection strategy and stacking method including hidden layer neural network. The combination of many methods helps to improve the accuracy of the model and the accuracy of blood sugar state prediction.

[0077] Furthermore, if Figure 4 As shown, based on the above-mentioned blood sugar state prediction method based on the ensemble learning model, the present invention also provides a blood sugar state prediction system based on the ensemble learning model, wherein the blood sugar state prediction system based on the ensemble learning model includes: The target model training module 51 is used to construct a data set, preprocess the data set to obtain a preprocessed data set, construct an integrated learning model based on the preprocessed data set through a preset integration strategy, and train the integrated learning model to obtain a target blood glucose state prediction model; The blood glucose state prediction module 52 is used to collect the target basic information and target routine blood test information of the user to be predicted, input the target basic information and the target routine blood test information into the target blood glucose state prediction model through encryption technology for prediction, and output the blood glucose state prediction result based on the integrated learning model; The reminder and notification module 53 is used to generate personalized reminders according to the blood sugar status prediction results based on the integrated learning model, and send the blood sugar status prediction results based on the integrated learning model and the personalized reminders to the user to be predicted, wherein the personalized reminders include screening suggestions and lifestyle guidance.

[0078] Furthermore, if Figure 5 As shown, based on the above-mentioned blood sugar status prediction method and system based on the integrated learning model, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 5Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0079] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a blood glucose state prediction program 40 based on an integrated learning model is stored on the memory 20, and the blood glucose state prediction program 40 based on the integrated learning model can be executed by the processor 10, thereby realizing the blood glucose state prediction method based on the integrated learning model in the present application.

[0080] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the blood glucose status prediction method based on the integrated learning model.

[0081] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface.

[0082] In one embodiment, when the processor 10 executes the blood glucose state prediction program 40 based on the integrated learning model in the memory 20, the following steps are implemented: Constructing a data set, preprocessing the data set to obtain a preprocessed data set, constructing an integrated learning model based on the preprocessed data set through a preset integration strategy, and training the integrated learning model to obtain a target blood glucose state prediction model; Collecting target basic information and target routine blood test information of the user to be predicted, inputting the target basic information and the target routine blood test information into the target blood glucose state prediction model through encryption technology for prediction, and outputting the blood glucose state prediction result based on the integrated learning model; A personalized reminder is generated according to the blood sugar status prediction result based on the ensemble learning model, and the blood sugar status prediction result based on the ensemble learning model and the personalized reminder are sent to the user to be predicted, wherein the personalized reminder includes screening suggestions and lifestyle guidance.

[0083] The constructing of the data set specifically includes: Collecting historical basic information and historical routine blood test information from the data system as original features, obtaining blood sugar states corresponding to the historical basic information and the historical routine blood test information as labels, and constructing an original data set based on the original features and the labels; De-identifying the data in the original data set, removing the data that reflects personal identity information, and performing access control and encryption on the removed data to obtain the data set; The historical basic information includes age information, gender information, height information and weight information, and the historical routine blood test information includes complete blood cell count information, liver function information, kidney function information and blood lipid information.

[0084] The preprocessing of the data set to obtain a preprocessed data set specifically includes: Using reference values ​​of indicators corresponding to gender and age to perform feature normalization processing on the data set, unifying the dimension and range of the features, and obtaining a first intermediate data set; Using the isolation forest algorithm to calculate the average path length of each data point in the first intermediate data set, calculating the anomaly score of each data point according to the average path length, taking the anomaly score higher than a preset threshold as an outlier, and removing the data points corresponding to the outlier to obtain a second intermediate data set; Generate new minority class samples using the SMOTE algorithm, synthesize the minority class samples with the second intermediate data set to obtain synthesized samples, clean the synthesized samples using the Tomek link technology, remove overlapping sample data and noise sample data on the boundary, and generate a third intermediate data set; The third intermediate data set is subjected to collinearity detection using a correlation coefficient matrix detection method and a variance inflation factor detection method, and the detected collinear variables are removed from the third intermediate data set to obtain a preprocessed data set.

[0085] Wherein, the preset integration strategy includes a multi-objective iterative model selection method and a stacking method including a hidden layer neural network; The step of constructing an integrated learning model based on the preprocessed data set through a preset integration strategy specifically includes: Constructing a candidate model set, and generating an integrated model through the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set; The integrated model is stacked by a stacking method including a hidden layer neural network to generate an integrated learning model of a stacked meta-learner.

[0086] The step of constructing a candidate model set, based on the preprocessed data set and the candidate model set, generating an integrated model through the multi-objective iterative model selection method, specifically includes: Acquire multiple initial candidate models, select hyperparameters for each of the initial candidate models using a Bayesian optimization algorithm to obtain candidate models, and construct a model candidate set including the multiple candidate models; Using the preprocessed data set to cross-validate each candidate model in the model candidate set, generating a prediction result for each candidate model, calculating the individual accuracy of each candidate model based on a plurality of the prediction results, and adding the candidate model with the highest individual accuracy from the model candidate set to the new model subset; Calculating the diversity of all remaining candidate models in the model candidate set relative to the model subset, normalizing the diversity and individual accuracy corresponding to each remaining candidate model, and obtaining the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model; Calculating the cost functions of all remaining candidate models according to the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model, and adding the remaining candidate model with the minimum cost function from the model candidate set to the model subset; Using the model subset to construct an initial integrated model, and using the preprocessed data set to cross-validate the integrated model, if the performance improvement of the initial integrated model does not exceed a preset threshold, the initial integrated model is used as the integrated model; If the performance improvement of the initial integrated model exceeds a preset threshold, continue to select the remaining candidate models with the minimum cost function from the model candidate set to add to the model subset until the performance of the initial integrated model no longer improves or the improvement is less than the preset improvement value, remove the remaining candidate model last added to the initial integrated model, and obtain the integrated model.

[0087] The stacking of the integrated model by a stacking method including a hidden layer neural network to generate an integrated learning model of a stacked meta-learner specifically includes: The preprocessed data set is divided into a training set and a test set according to a preset ratio, and each base learner in the integrated model is trained and predicted by a K-fold cross-validation algorithm according to the training set to obtain a first prediction result of each base learner on the training set, a primary feature matrix is ​​generated according to the first prediction result, and a second prediction result is generated using the test set; Wherein, each base learner in the integrated model corresponds to a model in the integrated model; Using the primary feature matrix as a training feature of a meta-learner, using the second prediction result as a test feature of the meta-learner, and using the label of the training set as a training label and a test label of the meta-learner to construct a meta-learner data set; Constructing an ensemble learning model, setting an input layer of the ensemble learning model, adding a preset number of meta-learners after the input layer, and setting an output layer of the ensemble learning model according to the task type; The number of nodes in the input layer is equal to the number of base learners, and each node corresponds to the first prediction result of a base learner.

[0088] The step of training the integrated learning model to obtain a target blood glucose state prediction model specifically includes: The ensemble learning model is trained and tested using the meta-learner dataset, and the prediction of the ensemble learning model is decomposed into multiple interpretable parts using Shapley additive explanation, each of which represents the contribution of a feature in the current sample to achieve an evaluation of the importance of the feature; The performance index of the integrated learning model is calculated, and if the performance index meets the preset conditions, a target blood glucose state prediction model is obtained.

[0089] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a blood glucose status prediction program based on an integrated learning model, and when the blood glucose status prediction program based on an integrated learning model is executed by a processor, the steps of the blood glucose status prediction method based on an integrated learning model as described above are implemented.

[0090] In summary, the present invention proposes a blood glucose status prediction method, system and terminal based on an integrated learning model, the method comprising: constructing a data set, preprocessing the data set to obtain a preprocessed data set, constructing an integrated learning model based on the preprocessed data set through an integrated strategy, training the integrated learning model to obtain a target blood glucose status prediction model; collecting target basic information and target routine blood test information of the user to be predicted, inputting the target basic information and target routine blood glucose status prediction model into the target blood glucose status prediction model for prediction, and outputting a blood glucose status prediction result based on the integrated learning model; generating personalized reminders according to the blood glucose status prediction results based on the integrated learning model, and sending the blood glucose status prediction results and personalized reminders to the user; the present invention innovatively proposes an integrated learning model that combines multiple data processing methods with integrated strategies, thereby improving the accuracy of blood glucose status prediction.

[0091] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0092] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0093] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for predicting blood sugar status based on an integrated learning model, characterized in that: The blood sugar state prediction method based on the integrated learning model includes: Constructing a data set, preprocessing the data set to obtain a preprocessed data set, constructing an integrated learning model based on the preprocessed data set through a preset integration strategy, and training the integrated learning model to obtain a target blood glucose state prediction model; Collecting target basic information and target routine blood test information of the user to be predicted, inputting the target basic information and the target routine blood test information into the target blood glucose state prediction model through encryption technology for prediction, and outputting the blood glucose state prediction result based on the integrated learning model; A personalized reminder is generated according to the blood sugar status prediction result based on the ensemble learning model, and the blood sugar status prediction result based on the ensemble learning model and the personalized reminder are sent to the user to be predicted, wherein the personalized reminder includes screening suggestions and lifestyle guidance.

2. The blood sugar status prediction method based on the ensemble learning model according to claim 1, characterized in that: The constructing of the data set specifically includes: Collecting historical basic information and historical routine blood test information from the data system as original features, obtaining blood sugar states corresponding to the historical basic information and the historical routine blood test information as labels, and constructing an original data set based on the original features and the labels; De-identifying the data in the original data set, removing the data that reflects personal identity information, and performing access control and encryption on the removed data to obtain the data set; The historical basic information includes age information, gender information, height information and weight information, and the historical routine blood test information includes complete blood cell count information, liver function information, kidney function information and blood lipid information.

3. The blood sugar status prediction method based on the ensemble learning model according to claim 1, characterized in that: The preprocessing of the data set to obtain a preprocessed data set specifically includes: Using reference values ​​of indicators corresponding to gender and age to perform feature normalization processing on the data set, unifying the dimension and range of the features, and obtaining a first intermediate data set; Using the isolation forest algorithm to calculate the average path length of each data point in the first intermediate data set, calculating the anomaly score of each data point according to the average path length, taking the anomaly score higher than a preset threshold as an outlier, and removing the data points corresponding to the outlier to obtain a second intermediate data set; Generate new minority class samples using the SMOTE algorithm, synthesize the minority class samples with the second intermediate data set to obtain synthesized samples, clean the synthesized samples using the Tomek link technology, remove overlapping sample data and noise sample data on the boundary, and generate a third intermediate data set; The third intermediate data set is subjected to collinearity detection using a correlation coefficient matrix detection method and a variance inflation factor detection method, and the detected collinear variables are removed from the third intermediate data set to obtain a preprocessed data set.

4. The method for predicting blood sugar status based on an integrated learning model according to claim 1, characterized in that: The preset integration strategy includes a multi-objective iterative model selection method and a stacking method including a hidden layer neural network; The step of constructing an integrated learning model based on the preprocessed data set through a preset integration strategy specifically includes: Constructing a candidate model set, and generating an integrated model through the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set; The integrated model is stacked by a stacking method including a hidden layer neural network to generate an integrated learning model of a stacked meta-learner.

5. The blood sugar status prediction method based on the ensemble learning model according to claim 4, characterized in that: The constructing of the candidate model set, generating an integrated model by the multi-objective iterative model selection method according to the preprocessed data set and the candidate model set, specifically includes: Acquire multiple initial candidate models, select hyperparameters for each of the initial candidate models using a Bayesian optimization algorithm to obtain candidate models, and construct a model candidate set including the multiple candidate models; Using the preprocessed data set to cross-validate each candidate model in the model candidate set, generating a prediction result for each candidate model, calculating the individual accuracy of each candidate model based on a plurality of the prediction results, and adding the candidate model with the highest individual accuracy from the model candidate set to the new model subset; Calculating the diversity of all remaining candidate models in the model candidate set relative to the model subset, normalizing the diversity and individual accuracy corresponding to each remaining candidate model, and obtaining the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model; Calculating the cost functions of all remaining candidate models according to the normalized diversity and normalized individual accuracy corresponding to each remaining candidate model, and adding the remaining candidate model with the minimum cost function from the model candidate set to the model subset; Using the model subset to construct an initial integrated model, and using the preprocessed data set to cross-validate the integrated model, if the performance improvement of the initial integrated model does not exceed a preset threshold, the initial integrated model is used as the integrated model; If the performance improvement of the initial integrated model exceeds a preset threshold, continue to select the remaining candidate models with the minimum cost function from the model candidate set to add to the model subset until the performance of the initial integrated model no longer improves or the improvement is less than the preset improvement value, remove the remaining candidate model last added to the initial integrated model, and obtain the integrated model.

6. The method for predicting blood sugar status based on an integrated learning model according to claim 5, characterized in that: The stacking method including a hidden layer neural network is used to stack the integrated model to generate an integrated learning model of a stacked meta-learner, specifically including: The preprocessed data set is divided into a training set and a test set according to a preset ratio, and each base learner in the integrated model is trained and predicted by a K-fold cross-validation algorithm according to the training set to obtain a first prediction result of each base learner on the training set, a primary feature matrix is ​​generated according to the first prediction result, and a second prediction result is generated using the test set; Wherein, each base learner in the integrated model corresponds to a model in the integrated model; Using the primary feature matrix as a training feature of a meta-learner, using the second prediction result as a test feature of the meta-learner, and using the label of the training set as a training label and a test label of the meta-learner to construct a meta-learner data set; Constructing an ensemble learning model, setting an input layer of the ensemble learning model, adding a preset number of meta-learners after the input layer, and setting an output layer of the ensemble learning model according to the task type; The number of nodes in the input layer is equal to the number of base learners, and each node corresponds to the first prediction result of a base learner.

7. The method for predicting blood sugar status based on an integrated learning model according to claim 6, characterized in that: The training of the integrated learning model to obtain a target blood glucose state prediction model specifically includes: The ensemble learning model is trained and tested using the meta-learner dataset, and the prediction of the ensemble learning model is decomposed into multiple interpretable parts using Shapley additive explanation, each of which represents the contribution of a feature in the current sample to achieve an evaluation of the importance of the feature; The performance index of the integrated learning model is calculated, and if the performance index meets the preset conditions, a target blood glucose state prediction model is obtained.

8. A blood sugar status prediction system based on an integrated learning model, characterized in that: The blood sugar state prediction system based on the integrated learning model includes: A target model training module is used to construct a data set, preprocess the data set to obtain a preprocessed data set, construct an integrated learning model based on the preprocessed data set through a preset integration strategy, and train the integrated learning model to obtain a target blood sugar state prediction model; A blood sugar status prediction module, which is used to collect the target basic information and target routine blood test information of the user to be predicted, input the target basic information and the target routine blood test information into the target blood sugar status prediction model through encryption technology for prediction, and output the blood sugar status prediction result based on the integrated learning model; A reminder and notification module is used to generate personalized reminders according to the blood sugar status prediction results based on the integrated learning model, and send the blood sugar status prediction results based on the integrated learning model and the personalized reminders to the user to be predicted, wherein the personalized reminders include screening suggestions and lifestyle guidance.

9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a blood glucose status prediction program based on an integrated learning model stored in the memory and executable on the processor. When the blood glucose status prediction program based on the integrated learning model is executed by the processor, the steps of the blood glucose status prediction method based on the integrated learning model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a blood glucose status prediction program based on an integrated learning model, and when the blood glucose status prediction program based on an integrated learning model is executed by a processor, the steps of the blood glucose status prediction method based on an integrated learning model as described in any one of claims 1-7 are implemented.

Images