Collaborative crowd health intervention method and system based on compliance prediction

Abstract

A health intervention method for healthy and healthy people based on compliance prediction relates to the technical field of health intervention, and mainly comprises the following steps: collecting training sample data to perform model pre-training on a deep belief network, collecting actual data of healthy and healthy people to be intervened, and sending the actual data to the pre-trained deep belief network to perform re-diagnosis compliance prediction; and outputting a re-visit compliance prediction classification result, adjusting and optimizing a health intervention strategy for the crowd according to the prediction classification result, and sending intervention knowledge content according to the intervention strategy and the patient condition. The invention further provides a health intervention system for healthy and healthy people based on compliance prediction. The system comprises a data synchronization module, a compliance prediction module, an intervention knowledge base, an intervention rule updating module and an intervention content sending module. According to the invention, a compliance prediction model for healthy and healthy people is constructed based on the deep belief network, and is used for guiding, adjusting and optimizing a health intervention strategy for people.

Description

Technical Field

[0001] The present invention relates to the technical field of health intervention, and particularly relates to a health intervention method and system for the elderly care population based on compliance prediction. Background Art

[0002] At present, the aging process in China is deepening continuously. The China Statistical Yearbook (2023) shows that the number of people aged 65 and above in China has reached 209.78 million, accounting for 14.9% of the total population. With the increase in the proportion of the elderly population, coupled with the hollowing out of families and the weakening of family care capabilities, there are requirements for the socialized elderly care to adjust the structure and emphasize services. The research on the "medical-care-elderly care" integrated elderly care service system and health support system for the elderly care population mainly composed of the elderly has important economic value and social significance.

[0003] The health intervention for the elderly care population usually includes multiple dimensions such as nutrition, exercise, medication, medical treatment, and psychology, and requires continuous and personalized intervention guidance according to individual differences and changes. Currently, some solutions have improved the work efficiency of medical staff by building an information system including an intervention knowledge base and an intervention rule base. However, the intervention rules do not consider the compliance differences and changes of the population, and use a relatively fixed triggering method, which on the one hand causes waste of valuable medical staff resources, and on the other hand also causes unnecessary interference to the elderly care population.

[0004] Specifically, the main disadvantages of the existing solutions are the health intervention rules, including the intervention frequency, content, sending objects, etc., are fixed or have low flexibility, resulting in the following problems:

[0005] (1) It cannot meet the needs of long-term management and intervention for the elderly care population. In the long run, people's health status, compliance level, etc. will change, and the intervention plan needs to be adjusted continuously.

[0006] (2) The intervention accuracy is not high.

[0007] (3) The adjustment of the intervention plan requires manual participation, and the value of medical staff resources is not fully utilized.

[0008] (4) It may cause excessive interference to the intervened population. For example, a person who does not understand a healthy lifestyle has been able to maintain a healthy lifestyle after continuous intervention in the early stage. However, if he is still continuously reminded by phone, text message, or WeChat to control smoking and drinking, it will be a nuisance to him. Summary of the Invention

[0009] Based on the above problems, the object of the present invention is to provide a health intervention method and system for the health care population based on compliance prediction. The method and system construct a compliance prediction model for the health care population based on a deep belief network, and consider time correlation to predict the compliance level of different populations in the region at the current time point, so as to guide the adjustment and optimization of the health intervention strategy for the population.

[0010] The technical solution adopted by the present invention to achieve its invention object is a health intervention method for the health care population based on compliance prediction, and its steps include:

[0011] S1. Collect training sample data of a predetermined number of health care populations and perform normalization preprocessing, and then send it to the Deep Belief Network in the compliance prediction module for model pre-training. The Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBMs) and one layer of Back Propagation (BP) module;

[0012] S2. Collect the actual data of the health care population to be intervened and perform normalization preprocessing, and then send it to the Deep Belief Network that has completed pre-training in the compliance prediction module;

[0013] S3. Use the Deep Belief Network that has completed pre-training in the compliance prediction module to process the actual data of the health care population to be intervened, perform follow-up visit compliance prediction, and output the follow-up visit compliance prediction classification result;

[0014] S4. According to the follow-up visit compliance prediction classification result obtained in step S3, adjust and optimize the health intervention strategy for this population;

[0015] S5. Send intervention knowledge content according to the intervention strategy and the patient's situation.

[0016] Further, the specific steps of the model pre-training in step S1 are:

[0017] S1.1. Construct the parameter space of the first-layer Restricted Boltzmann Machine (RMB), that is, determine the hyperparameters including at least the learning rate, the number of iterations, and the network structure parameters;

[0018] S1.2. The preprocessed training sample data after normalization enters the first-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training;

[0019] S1.3. The feature vector extracted after the training of the first-layer Restricted Boltzmann Machine (RBM) is used as the input to enter the next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training;

[0020] S1.4. The feature vectors extracted after the training of the lower-layer restricted Boltzmann machine (RBM) described in step S1.3 are continued as inputs to the next lower-layer restricted Boltzmann machine (RBM) to complete unsupervised training, and the iteration continues until the last-layer restricted Boltzmann machine (RBM) completes unsupervised training. At this time, through the unsupervised training of each layer of restricted Boltzmann machines (RBMs), the current optimal parameters of each layer are obtained.

[0021] S1.5. Perform supervised training through the backpropagation (BP) module. The error obtained from the supervised training is backpropagated to each layer of restricted Boltzmann machines (RBMs), and the parameter weights of each layer of restricted Boltzmann machines (RBMs) are adjusted to obtain the current global optimal parameters of the deep belief network (DBN), thus completing the pre-training of the model.

[0022] Further, the specific steps of step S3 are as follows:

[0023] S3.1. The actual data of the health care population to be intervened after normalization preprocessing enters the first-layer restricted Boltzmann machine (RBM) to complete unsupervised training.

[0024] S3.2. The feature vectors extracted after the training of the first-layer restricted Boltzmann machine (RBM) are used as inputs to the next lower-layer restricted Boltzmann machine (RBM) to complete unsupervised training.

[0025] S3.3. The feature vectors extracted after the training of the next lower-layer restricted Boltzmann machine (RBM) described in step S3.2 are continued as inputs to the next lower-layer restricted Boltzmann machine (RBM) to complete unsupervised training, and the iteration continues until the last-layer restricted Boltzmann machine (RBM) completes unsupervised training.

[0026] S3.4. Perform supervised training through the backpropagation (BP) module. The error obtained from the supervised training is backpropagated to each layer of restricted Boltzmann machines (RBMs), and the parameter weights of each layer of restricted Boltzmann machines (RBMs) are adjusted to obtain and output the predicted data of follow-up compliance.

[0027] S3.5. Classify the predicted data of follow-up compliance through a Softmax classifier to obtain the classification result of the predicted follow-up compliance.

[0028] Further, the classification result of the predicted follow-up compliance includes four discrete values: excellent, good, medium, and poor.

[0029] Further, the training sample data of the health care population and the actual data of the health care population to be intervened include continuous real-valued data, binary classification data, and finite data sequences, where:

[0030] The continuous real-valued data includes at least blood pressure, blood sugar, glycated hemoglobin, and blood lipid measurement values.

[0031] The binary classification data at least includes whether smoking, whether there is a family history, whether there is a diagnosis prescription, whether there are two or more diagnosis prescriptions, whether one can maintain weekly exercise, and whether one is an urban resident;

[0032] The finite data sequence at least includes the month and the number of follow - ups received within the last 6 months.

[0033] Furthermore, the specific method for normalizing and pre - processing the training sample data of the health - care population and the actual data of the health - care population to be intervened is as follows:

[0034] For continuous real - valued data, use Z - score for normalization processing to map the result value to between 0 and 1;

[0035] For binary classification data, adopt binary encoding, where 1 represents yes and 0 represents no;

[0036] Encode the finite data sequence between 0 and 1.

[0037] Further, the specific method for sending the intervention knowledge content in step S5 is as follows:

[0038] If the intervention form output by the intervention strategy is information reach, then judge whether the user logs in through WeChat. If logged in, push the intervention knowledge content to the user in the form of a WeChat template message. If not logged in, do not push the WeChat template message, and send the intervention knowledge content to the user or their family members via text message;

[0039] If the intervention form output by the intervention strategy is phone reach, then the system automatically generates an intervention follow - up task and pushes it to the operation terminal of the responsible medical staff. After the responsible medical staff completes the intervention task, the system retains the intervention path and results.

[0040] On the other hand, the present invention also provides a health - care population health intervention system based on compliance prediction, including a data synchronization module, a compliance prediction module, an intervention knowledge base, an intervention rule update module, and an intervention content sending module, where:

[0041] The data synchronization module is used to collect the training sample data of the health - care population or the actual data of the health - care population to be intervened and perform normalization pre - processing;

[0042] The compliance prediction module is used to receive the training sample data of the health - care population after normalization pre - processing and perform model pre - training of the Deep Belief Network, and is also used to receive the actual data of the health - care population to be intervened after normalization pre - processing and use the pre - trained Deep Belief Network to predict the follow - up compliance, and output the follow - up compliance prediction classification result;

[0043] The intervention knowledge base is used to store intervention knowledge content;

[0044] The intervention rule update module is used to adjust and optimize the health intervention strategy for this population according to the follow-up compliance prediction classification result, and is used to call the intervention knowledge content stored in the intervention knowledge base and send it to the intervention content sending module;

[0045] The intervention content sending module is used to send the intervention knowledge content according to the intervention strategy and the patient's situation.

[0046] Furthermore, the Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBM) and one layer of Back Propagation (BP) module.

[0047] The beneficial effects of the present invention are as follows:

[0048] (1) The present invention has a compliance prediction module based on the Deep Belief Network, which can mine and analyze the compliance grouping of users through the statistical characteristics of the population data in this region and the user's historical data, and then guide the adjustment of the intervention plan.

[0049] (2) The present invention has an intervention rule update module, which can dynamically adjust the precise health intervention plan for the elderly care population. On the premise of ensuring the intervention effects such as the follow-up rate and the index control rate, it reduces the interference to users and improves the work efficiency of medical staff.

[0050] (3) When the Deep Belief Network of the present invention is used for compliance prediction, it can fully consider the influence of different regions and time seasons, add regional and time factors to the data dimension, and improve the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the overall process of the embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram of the pre-training process of the Deep Belief Network model in the embodiment of the present invention;

[0053] Figure 3 It is a schematic diagram of the compliance prediction process using the pre-trained Deep Belief Network in the embodiment of the present invention;

[0054] Figure 4 It is a schematic diagram of the system structure in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0056] Embodiment 1

[0057] Figures 1 to 3 The first specific implementation manner of the present invention is shown, which is a health intervention method for the elderly and convalescent population based on compliance prediction. Its steps include:

[0058] S1. The data synchronization module collects training sample data of a predetermined number of the elderly and convalescent population and performs normalization preprocessing, and then sends it to the Deep Belief Network for model pre-training. The Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBMs) and one layer of Back Propagation (BP) module;

[0059] S2. The data synchronization module collects the actual data of the elderly and convalescent population to be intervened and performs normalization preprocessing, and then sends it to the Deep Belief Network that has completed pre-training;

[0060] S3. Use the Deep Belief Network that has completed pre-training to process the actual data of the elderly and convalescent population to be intervened, perform follow-up compliance prediction, and output the follow-up compliance prediction classification result; the follow-up compliance prediction classification result includes four discrete values: excellent, good, medium, and poor.

[0061] S4. The intervention rule update module adjusts and optimizes the health intervention strategy for this population according to the follow-up compliance prediction classification result obtained in step S3, and calls the intervention knowledge content stored in the intervention knowledge base and sends it to the intervention content sending module;

[0062] Predict and classify the follow-up compliance of users, aiming to guide the adjustment and optimization of the health intervention strategy for this population. For users with poor compliance, perform strong intervention, increase the intervention frequency, select richer intervention content, and synchronize the intervention content to the user's family members or responsible medical staff. For users with good compliance, reduce the intervention frequency and only push the content to the users at important time points. For example, for the health intervention strategy for type 2 diabetes patients, continue to perform strong intervention follow-up on patients with a prediction result of "poor". The intervention benchmark within one follow-up cycle includes 9 intervention contents, as shown in the following table. If the compliance prediction result of the user is "excellent", cancel the third, fourth, and eighth intervention contents, and only perform 6 interventions within one intervention cycle.

[0063]

[0064] S5. The intervention content sending module sends the intervention knowledge content according to the intervention strategy and the patient's situation.

[0065] The intervention knowledge base stores intervention knowledge content in multiple dimensions, including nutrition, exercise, medication, medical treatment, psychology, etc. Medical experts compile it according to content classification, and the forms include short sentences, articles, videos, passwords, Q&A, etc. Taking the pharmaceutical service library of drugs as an example, it can include usage and dosage, drug knowledge, disease knowledge, medication compliance education, etc.

[0066] In this embodiment, the specific steps of the model pre-training in step S1 are as follows:

[0067] S1.1. Construct the parameter space of the first-layer restricted Boltzmann machine RMB, that is, determine the hyperparameters including at least the learning rate, the number of iterations, and the network structure parameters;

[0068] S1.2. The preprocessed and normalized training sample data enters the first-layer restricted Boltzmann machine RBM to complete unsupervised training;

[0069] S1.3. The feature vectors extracted after the training of the first-layer restricted Boltzmann machine RBM are used as inputs to enter the next-layer restricted Boltzmann machine RBM to complete unsupervised training;

[0070] S1.4. The feature vectors extracted after the training of the next-layer restricted Boltzmann machine RBM in step S1.3 continue to be used as inputs to enter the next-next-layer restricted Boltzmann machine RBM to complete unsupervised training, and the iteration continues until the last-layer restricted Boltzmann machine RBM completes unsupervised training; at this time, through the unsupervised training of each layer of restricted Boltzmann machine RBM, the current optimal parameters of each layer are obtained;

[0071] S1.5. Perform supervised training through the backpropagation module BP. The error obtained from the supervised training is backpropagated to each layer of restricted Boltzmann machine RBM, and the parameter weights of each layer of restricted Boltzmann machine RBM are adjusted to obtain the current global optimal parameters of the deep belief network Deep Belief Network, and the model pre-training is completed.

[0072] The deep belief network Deep Belief Network is a neural network model including multiple hidden layers, including multiple layers of restricted Boltzmann machine RBM and one layer of backpropagation module BP;

[0073] Its energy function can be expressed as:

[0074]

[0075] where v and h represent the visible layer and the hidden layer respectively, a and b are the offsets of the visible layer and the hidden layer respectively, W is the weight matrix between the visible layer and the hidden layer, and the model parameters of RBM are expressed as

[0076] θ = {W ij ,ai , b j}。

[0077] The joint probability distribution of the visible layer and the hidden layer is as follows:

[0078]

[0079] where is the normalization factor.

[0080] Since there is no direct connection between neurons in each layer of the RBM, and the conditional distributions of the neuron states in each layer are independent of each other, the probability expressions for the activation states of the visible layer and the hidden layer neurons can be obtained as follows:

[0081]

[0082]

[0083] where the activation function is the sigmoid function.

[0084] The input data vector is input from the bottom of the Deep Belief Network, and is subjected to feature extraction training layer by layer through multiple stacked Restricted Boltzmann Machines (RBMs). A backpropagation (BP) layer is stacked on the top layer, and finally data output is achieved. The whole process is equivalent to first obtaining the optimal solution in the model area, and then further screening out the optimal solution in the optimal solution area.

[0085] In this embodiment, the specific steps of step S3 are as follows:

[0086] S3.1. The actual data of the rehabilitation population to be intervened after normalization preprocessing enters the first-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training;

[0087] S3.2. The feature vector extracted after the training of the first-layer Restricted Boltzmann Machine (RBM) is used as input to enter the next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training;

[0088] S3.3. The feature vector extracted after the training of the next-layer Restricted Boltzmann Machine (RBM) described in step S3.2 continues to be used as input to enter the next-next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training, and the iteration continues until the last-layer Restricted Boltzmann Machine (RBM) completes unsupervised training;

[0089] S3.4. Perform supervised training through the backpropagation module BP. The error obtained from the supervised training is backpropagated to each layer of the Restricted Boltzmann Machine (RBM), and the parameter weights of each layer of the Restricted Boltzmann Machine (RBM) are adjusted to obtain and output the predicted data of follow-up compliance;

[0090] S3.5. Classify the follow-up compliance prediction data through a Softmax classifier to obtain the follow-up compliance prediction classification result.

[0091] The follow-up compliance prediction classification result includes four discrete values: excellent, good, medium, and poor.

[0092] In this embodiment, the data synchronization module grabs data including hospital information system data, public health data, patient wearable device detection data, patient scale feedback data, etc. through a standard data interface, realizes multi-dimensional data capture and processing of the elderly care population, uniformly aggregates and correlates and analyzes data from various sources, and can form an accurate population label portrait. The elderly care population training sample data and the actual data of the elderly care population to be intervened include continuous real-valued data, binary classification data, and finite data sequences, where:

[0093] The continuous real-valued data includes at least blood pressure, blood sugar, glycated hemoglobin, and blood lipid measurement values;

[0094] The binary classification data includes at least whether smoking, whether there is a family history, whether there is a diagnosis prescription, whether there are two or more diagnosis prescriptions, whether one can maintain weekly exercise, and whether one is an urban resident;

[0095] The finite data sequence includes at least month and the number of follow-up visits received within the last 6 months. For example, the month takes the natural month values from 1 to 12, and the influence of different times due to seasons, temperatures, festivals, special government policies, etc. on population compliance can be considered.

[0096] The specific method for performing normalization preprocessing on the elderly care population training sample data and the actual data of the elderly care population to be intervened is as follows:

[0097] For continuous real-valued data, use Z-score for normalization processing to map the result value to between 0 and 1;

[0098] For binary classification data, adopt binary coding, where 1 represents yes and 0 represents no;

[0099] Encode the finite data sequence between 0 and 1. For example, the months from 1 to 12 are encoded as {0, 0.09, 0.18,..., 0.9, 1}, and the number of follow-up visits received within the last 6 months is encoded as {0, 0.1, 0.2,..., 0.9, 1} (not receiving a follow-up visit is encoded as 0, receiving 1 follow-up visit is encoded as 0.1, receiving 2 follow-up visits is encoded as 0.2,..., receiving 10 or more follow-up visits is encoded as 1).

[0100] In this embodiment, the specific method for sending the intervention knowledge content in step S5 is as follows:

[0101] If the intervention form output by the intervention strategy is information dissemination, it is judged whether the user logs in through WeChat. If logged in, the intervention knowledge content is pushed to the user in the form of a WeChat template message. If not logged in, the WeChat template message is not pushed, and the intervention knowledge content is sent to the user or their family members via SMS;

[0102] If the intervention form output by the intervention strategy is phone call, the system automatically generates an intervention follow-up task and pushes it to the operation terminal of the responsible medical staff. After the responsible medical staff completes the intervention task, the intervention path and results are retained by the system.

[0103] Embodiment 2

[0104] Figure 4 The second specific implementation manner of the present invention is shown, including a data synchronization module, a compliance prediction module, an intervention knowledge base, an intervention rule update module, and an intervention content sending module, where:

[0105] The data synchronization module is used to collect the training sample data of the elderly care population or the actual data of the elderly care population to be intervened and perform normalization preprocessing;

[0106] The compliance prediction module is used to receive the training sample data of the elderly care population after normalization preprocessing and perform model pre-training of the Deep Belief Network, and is also used to receive the actual data of the elderly care population to be intervened after normalization preprocessing and use the pre-trained Deep Belief Network to predict the follow-up visit compliance, and output the follow-up visit compliance prediction classification result;

[0107] The intervention knowledge base is used to store the intervention knowledge content;

[0108] The intervention rule update module is used to adjust and optimize the health intervention strategy for this population according to the follow-up visit compliance prediction classification result, and is used to call the intervention knowledge content stored in the intervention knowledge base to the intervention content sending module;

[0109] The intervention content sending module is used to send the intervention knowledge content according to the intervention strategy and the patient's situation.

[0110] In this embodiment, the Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBM) and one layer of Back Propagation (BP) module.

[0111] In the present invention, each implementation manner can be realized in the way of software plus a necessary hardware platform.

[0112] The above embodiments of the present invention are merely examples for illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes and modifications can be made on the basis of the above description. It is impossible to list all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A health intervention method for the health care population based on compliance prediction, characterized in that Including the following steps: S1. Collect training sample data of a predetermined number of health care population and perform normalization preprocessing, and then send it into the Deep Belief Network for model pre-training. The Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBMs) and one layer of Back Propagation (BP) module; S2. Collect the actual data of the health care population to be intervened and perform normalization preprocessing, and then send it to the Deep Belief Network that has completed pre-training; S3. Use the Deep Belief Network that has completed pre-training to process the actual data of the health care population to be intervened, perform follow-up compliance prediction, and output the follow-up compliance prediction classification result; S4. According to the follow-up compliance prediction classification result obtained in step S3, adjust and optimize the health intervention strategy for this population; S5. Send intervention knowledge content according to the intervention strategy and the patient's situation.

2. The health intervention method for the elderly and convalescent people based on compliance prediction according to claim 1, characterized in that The specific steps of the model pre-training in step S1 are as follows: S1.

1. Construct the parameter space of the first-layer Restricted Boltzmann Machine (RMB), that is, determine the hyperparameters including at least the learning rate, the number of iterations, and the network structure parameters; S1.

2. The preprocessed training sample data after normalization enters the first-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training; S1.

3. The feature vector extracted after the training of the first-layer Restricted Boltzmann Machine (RBM) is used as the input to enter the next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training; S1.

4. The feature vector extracted after the training of the next-layer Restricted Boltzmann Machine (RBM) described in step S1.3 continues to be used as the input to enter the next-next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training, and the iteration continues until the last-layer Restricted Boltzmann Machine (RBM) completes unsupervised training; At this time, through the unsupervised training of each layer of Restricted Boltzmann Machines (RBMs), the current optimal parameters of each layer are obtained; S1.

5. Perform supervised training through the Back Propagation (BP) module. The error obtained from the supervised training is backpropagated to each layer of Restricted Boltzmann Machines (RBMs), and the parameter weights of each layer of Restricted Boltzmann Machines (RBMs) are adjusted to obtain the current global optimal parameters of the Deep Belief Network, and the model pre-training is completed.

3. The health intervention method for the health care population based on compliance prediction according to claim 1, wherein, The specific steps of step S3 are as follows: S3.

1. The preprocessed actual data of the health care population to be intervened enters the first-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training; S3.

2. The feature vector extracted after the training of the first-layer Restricted Boltzmann Machine (RBM) is used as the input to enter the next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training; S3.

3. The feature vector extracted after the training of the next-layer Restricted Boltzmann Machine (RBM) described in step S3.2 continues to be used as the input to enter the next-next-layer Restricted Boltzmann Machine (RBM) to complete unsupervised training, and the iteration continues until the last-layer Restricted Boltzmann Machine (RBM) completes unsupervised training; S3.

4. Perform supervised training through the backpropagation module BP. The error obtained from the supervised training is backpropagated to each layer of the restricted Boltzmann machine (RBM), and the parameter weights of each layer of the RBM are adjusted to obtain and output the follow-up compliance prediction data. S3.

5. Classify the follow-up compliance prediction data through a Softmax classifier to obtain the follow-up compliance prediction classification result.

4. The health intervention method for the elderly population based on compliance prediction according to claim 1, wherein, The follow-up compliance prediction classification result includes four discrete values: excellent, good, medium, and poor.

5. The health intervention method for the elderly and convalescent people based on compliance prediction according to claim 1, wherein The training sample data of the health care population and the actual data of the health care population to be intervened include continuous real-valued data, binary classification data, and finite data sequences, where: The continuous real-valued data includes at least blood pressure, blood sugar, glycated hemoglobin, and blood lipid measurement values. The binary classification data includes at least whether the person smokes, whether there is a family history, whether there is a diagnosis prescription, whether there are two or more diagnosis prescriptions, whether they can maintain weekly exercise, and whether they are urban residents. The finite data sequence includes at least the month and the number of follow-up visits received within the last 6 months.

6. The health intervention method for the elderly care population based on compliance prediction according to claim 5, characterized in that, The specific method for normalizing and preprocessing the training sample data of the health care population and the actual data of the health care population to be intervened is as follows: For continuous real-valued data, use Z-score for normalization to map the result value to between 0 and 1. For binary classification data, use binary encoding, where 1 represents yes and 0 represents no. Encode the finite data sequence between 0 and 1.

7. A health intervention method for the elderly population based on compliance prediction according to claim 1, characterized in that The specific method for sending the intervention knowledge content in step S5 is as follows: If the intervention form output by the intervention strategy is information reach, determine whether the user has logged in to WeChat. If logged in, push the intervention knowledge content to the user in the form of a WeChat template message. If not logged in, do not push the WeChat template message and send the intervention knowledge content to the user or their family member via text message. If the intervention form output by the intervention strategy is phone reach, the system automatically generates an intervention follow-up task and pushes it to the operation terminal of the responsible medical staff. After the responsible medical staff completes the intervention task, the intervention path and result are retained by the system.

8. A health intervention system for the elderly and convalescent population based on compliance prediction, characterized in that, It includes a data synchronization module, a compliance prediction module, an intervention knowledge base, an intervention rule update module, and an intervention content sending module, where: The data synchronization module is used to collect the training sample data of the health care population or the actual data of the health care population to be intervened and perform normalization preprocessing. The compliance prediction module is used to receive the training sample data of the health care population after normalization preprocessing and perform model pre-training of the Deep Belief Network (DBN). It is also used to receive the actual data of the health care population to be intervened after normalization preprocessing and use the pre-trained DBN to predict the follow-up compliance, and output the follow-up compliance prediction classification result. The intervention knowledge base is used to store the intervention knowledge content. The intervention rule update module is used to adjust and optimize the health intervention strategy for this population according to the follow-up compliance prediction classification result, and is used to call the intervention knowledge content stored in the intervention knowledge base and send it to the intervention content sending module. The intervention content sending module is used to send intervention knowledge content according to the intervention strategy and the patient's condition.

9. The health intervention for the elderly and convalescent population based on compliance prediction according to claim 8, wherein: The Deep Belief Network includes multiple layers of Restricted Boltzmann Machines (RBM) and one layer of Back Propagation (BP) module.

Images