Remote monitoring and early warning system for heart failure patients based on internet of things

Abstract

The application discloses a heart failure patient remote monitoring and early warning system based on the Internet of Things, comprising a data acquisition and preprocessing module, which is used for acquiring multi-source physiological and behavioral data and preprocessing to generate an input sample set; a feature evaluation module, which is used for constructing an extreme gradient boosting tree algorithm model and obtaining a feature importance score; a feature screening module, which is used for executing a recursive feature elimination process and outputting an optimal feature subset; a risk discrimination module, which is used for inputting the optimal feature subset into the extreme gradient boosting tree algorithm model to generate a heart failure risk score; an early warning triggering module, which is used for triggering early warning and sending early warning information to the patient and the platform; a feedback recording module, which is used for generating feedback information; and a dynamic optimization module, which is used for updating the input sample set based on the feedback information, retraining the extreme gradient boosting tree algorithm model and screening features. The application combines the extreme gradient boosting tree and the recursive feature elimination, realizes precise evaluation and dynamic early warning of multi-source data of heart failure patients.

Description

Technical Field

[0001] This invention relates to the fields of medical and health monitoring and artificial intelligence technology, and in particular to a remote monitoring and early warning system for heart failure patients based on the Internet of Things. Background Technology

[0002] Currently, with the increasing aging population and the continuous growth in the number of heart failure patients, effective monitoring and early warning of chronic heart failure has become a core issue in clinical medicine and family health management. Traditional heart failure risk assessment relies heavily on manual intervention and regular checkups in hospitals, which struggles to capture dynamic changes in patients' vital signs in a timely manner and suffers from significant limitations such as data lag, slow response, and lack of continuity. With the development of IoT technology, more and more wearable devices and health sensors are being used for vital sign monitoring, making it possible to collect multi-source physiological and behavioral data such as heart rate, blood pressure, weight, sleep, and activity in patients' home environments, providing a new technological foundation for remote dynamic management of heart failure.

[0003] Existing research has attempted to apply machine learning methods to predict heart failure risk, such as using algorithms like logistic regression, support vector machines, or random forests to build scoring models. However, these models often face problems such as feature redundancy, poor model stability, and weak generalization ability when dealing with high-dimensional time-series data. Especially in multi-source IoT monitoring data, redundant features can not only lead to model overfitting but may also mask key variables crucial for risk assessment, severely affecting the accuracy and timeliness of risk prediction. Furthermore, existing methods generally lack adaptive optimization mechanisms for dynamic changes in the importance of model features, making it difficult to achieve automatic maintenance and continuous optimization of model performance over long-term operation. Simultaneously, most systems have failed to establish effective feedback learning mechanisms, failing to use historical warning results and actual processing effects for model correction and sample updates, resulting in models that struggle to adapt to long-term changes in patient conditions and individual differences.

[0004] Therefore, how to provide a remote monitoring and early warning system for heart failure patients based on the Internet of Things is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a remote monitoring and early warning system for heart failure patients based on the Internet of Things (IoT). This invention fully utilizes IoT multi-source health data acquisition technology and ensemble learning modeling methods, and describes in detail the entire process mechanism through feature importance assessment, feature screening, risk scoring, multi-level early warning, and dynamic feedback optimization. It has advantages such as high assessment accuracy, strong feature screening adaptability, continuous model optimization, and timely early warning response.

[0006] The Internet of Things-based remote monitoring and early warning system for heart failure patients according to an embodiment of the present invention includes the following modules:

[0007] The data acquisition and preprocessing module is used to collect multi-source physiological and behavioral data and preprocess them to generate an input sample set.

[0008] The feature evaluation module is used to build and train an extreme gradient boosting tree algorithm model and obtain feature importance scores.

[0009] The feature selection module is used to execute the recursive feature elimination process and output the optimal feature subset;

[0010] The risk assessment module is used to input the optimal feature subset into the extreme gradient boosting tree algorithm model to generate a heart failure risk score.

[0011] The early warning triggering module is used to trigger an early warning and send warning information to the patient and the platform based on the comparison between the heart failure risk score and the preset risk threshold.

[0012] The feedback recording module is used to record the warning score, warning level and platform response results, and generate feedback information;

[0013] The dynamic optimization module is used to update the input sample set based on feedback information, periodically retrain the extreme gradient boosting tree algorithm model and select features.

[0014] Optionally, modules can be integrated using the following methods:

[0015] S1. Collect multi-source physiological and behavioral data of heart failure patients through IoT devices, preprocess the multi-source physiological and behavioral data, and construct a structured input sample set;

[0016] S2. Construct an extreme gradient boosting tree algorithm model, train it using the input sample set, and obtain the feature importance score corresponding to each input feature;

[0017] S3. Based on the feature importance score, execute the recursive feature elimination process. In each iteration, remove the feature with the lowest score and retrain the extreme gradient boosting tree algorithm model until the set optimal feature subset size or performance convergence condition is met, and output the optimal feature subset.

[0018] S4. Based on the optimal feature subset, the input sample set is discriminated and the heart failure risk score within the corresponding time window is output. Multi-level risk thresholds are set according to the heart failure risk score. When any heart failure risk score exceeds the set risk threshold, the early warning process is triggered, an early warning signal is output and an early warning information is sent to the patient terminal device and the medical platform. At the same time, the early warning event and the processing result are recorded as feedback information.

[0019] S5. Compare the feedback information with the heart failure risk score, update the input sample set, and periodically re-execute the feature evaluation and screening process described in steps S2 and S3 to form a dynamically optimized risk scoring mechanism.

[0020] Optionally, the multi-source physiological and behavioral data specifically includes individual vital signs and behavioral parameters data collected through Internet of Things (IoT) devices, such as heart rate, blood pressure, respiratory rate, blood oxygen saturation, body temperature, weight, heart rate variability, sleep state, and body movement level.

[0021] Optionally, the preprocessing of multi-source physiological and behavioral data specifically includes performing data cleaning, missing value imputation, normalization, sliding window segmentation, and feature format unification.

[0022] Optionally, S2 specifically includes:

[0023] S21, The input sample set is ,in Indicates the first The original feature vector of each sample, This indicates the corresponding heart failure risk label. The number of samples;

[0024] S22. Construct a graph structure relationship diagram among input features. , where the set of nodes Represents input features, edge set The edges represent the correlation between features, and the edge weights are calculated based on the Pearson correlation coefficient between the features.

[0025] S23. Utilize a graph convolutional network to perform graph structure embedding processing on each sample in the structured input sample set, generating graph structure embedding vectors that fuse feature dependencies. ;

[0026] S24. Concatenate the graph structure embedding vector with the original feature vector to form the enhanced input vector. And construct an enhanced input sample set. ;

[0027] S25. Based on the enhanced input sample set, an extreme gradient boosting tree algorithm model is constructed. The extreme gradient boosting tree algorithm model consists of multiple regression trees. The enhanced input vector is used as the input of the extreme gradient boosting tree algorithm model. Through iterative training, the output of each regression tree is continuously accumulated to gradually approach the heart failure risk label value of the corresponding sample.

[0028] S26. Define a third-order approximation objective function based on the enhanced input vector. ;

[0029] S27. Constructing the structure regularization term in the objective function ;

[0030] S28. Set the initial predicted values ​​of all samples as a constant term. And set the initial learning rate to ;

[0031] S29, in the During the training rounds, the current model residual descent rate is taken into account. Cumulative forecast error volatility and the dynamic range width of the previous round of heart failure risk score results. Dynamically calculate adaptive learning rate During the training of the extreme gradient boosting tree algorithm model, the update step size is adaptively controlled.

[0032] S210. Based on the objective function and the structural regularization term, a greedy splitting strategy is used to construct a regression tree in the current round. and with adaptive learning rate Update the prediction output of the current extreme gradient boosting tree algorithm model;

[0033] S211. Repeat steps S25 to S210 until the extreme gradient boosting tree algorithm model reaches the set upper limit of the number of regression trees or the objective function satisfies the convergence condition.

[0034] S212. Based on the gain contribution of each input feature to the objective function in the trained extreme gradient boosting tree algorithm model, calculate the feature importance score for each input feature. Output feature importance score set , where M is the feature dimension.

[0035] Optionally, S3 specifically includes:

[0036] S31, Receive input sample set ,in Indicates the first The original feature vector of each sample, This indicates the corresponding heart failure risk label. The number of samples;

[0037] S32. Based on the set of feature importance scores output. Introducing a feature distribution consistency weight coefficient Calculate the weighted score vector The feature distribution consistency weight coefficient is obtained by calculating and normalizing the distribution differences of features in each sub-sample set.

[0038] ;

[0039] Where M is the feature dimension;

[0040] S33, Based on weighted scoring vector Sort in ascending order and select the one with the lowest score. 1 feature, construct a feature index set to be removed. From the input sample set The features corresponding to the feature index set to be removed are removed from the sample set to construct the first round of feature subset. ;

[0041] S34. The first round of feature subset sample set The input is fed into an extreme gradient boosting tree algorithm model for training, and the corresponding feature importance score vector is output. ;

[0042] S35. Set of feature indices to be removed Each feature in Calculate the marginal impact on the objective function of the extreme gradient boosting tree algorithm model. ;

[0043] S36. Let the current iteration number be... Based on the previous round of feature subset sample set and feature importance score vector Perform the following steps:

[0044] Calculate the feature retention score based on the marginal impact. ;

[0045] By combining feature retention scores and feature distribution consistency weighting coefficients, a tiered elimination strategy is implemented to construct a new set of feature subset samples. ;

[0046] The new round of feature subset sample set The input is fed into an extreme gradient boosting tree algorithm model for training, and the corresponding feature importance score vector is output. ;

[0047] Calculate the performance metrics of the extreme gradient boosting tree algorithm model in this round. ;

[0048] S37. Determine if the iteration termination condition is met. If it is, terminate the recursive feature elimination process; otherwise, let... Return to step S36 and continue execution. The iteration termination conditions include:

[0049] Input dimension in the current feature subset sample set Less than the set minimum feature dimension ;

[0050] The performance degradation of the extreme gradient boosting tree algorithm model meets the following requirements: ,in This is the performance tolerance threshold;

[0051] S38. Denote the iteration termination round as... Extract the currently retained set of feature indices. , as the final optimal feature subset;

[0052] S39. Based on the feature index set For structured input sample sets Perform feature mapping on all samples to generate the final input sample set. .

[0053] Optionally, S4 specifically includes:

[0054] S41. Receive the final input sample set. ,in, Indicates the first The input feature vector of each sample under the selected optimal feature subset dimension. This indicates the corresponding heart failure risk label. The number of samples;

[0055] S42. Calculate the optimal feature vector of each sample in the final input sample set. The input is fed into the constructed and trained extreme gradient boosting tree algorithm model to perform the classification and discrimination process for heart failure risk, and outputs the heart failure risk score for each sample within a set time window. This constitutes the heart failure risk score result;

[0056] S43. Construct a risk score vector from the heart failure risk scores corresponding to all samples. ,in Indicates the first The heart failure risk score results of each sample in the classification and discrimination process;

[0057] S44. Set a multi-level risk threshold set Among them, satisfying , Indicates the first Each risk level corresponds to a lower limit threshold for scoring, and a total of [number] thresholds are constructed. One risk level range;

[0058] S45. For any score in the risk scoring vector Determine the risk threshold set within the established multi-level risk threshold range. Within the risk level range to which it belongs, there is a unique risk level label. Satisfying the inequality and label the risk level As the first Output of heart failure risk level for each sample;

[0059] S46. Assemble a risk level label vector from all the risk level labels corresponding to the samples. ,in Represents the input feature vector The corresponding heart failure risk level result;

[0060] S47. Set the risk level threshold for triggering early warnings. When any risk level label If this occurs, immediately trigger the early warning process and initiate the heart failure risk emergency response procedure;

[0061] S48. Form an early warning trigger sample set by combining the sample indices that meet the early warning conditions. And targeting the early warning trigger sample set For each sample, the corresponding patient terminal device and medical service platform send an early warning signal and early warning information package containing risk level labels, risk score values ​​and time window information;

[0062] S49. Record the sample number corresponding to each warning event. Heart failure risk score Risk level label The system sends the patient's terminal device number and the early warning response processing information returned by the medical platform to construct feedback information. .

[0063] Optionally, S5 specifically includes:

[0064] S51. Extract the sample index of historical early warning events, heart failure risk score results, risk level labels and actual response and processing results from the feedback information to construct a feedback sample set;

[0065] S52. Compare the scoring results in the feedback sample set with the corresponding feedback information, analyze the accuracy of the extreme gradient boosting tree algorithm model scoring and the rationality of risk level classification, and identify samples and feature attributes whose scoring deviation is greater than the preset deviation threshold.

[0066] S53. Based on the comparison analysis results, the relevant sample labels in the original input sample set are corrected, and the newly added feedback samples and their latest label information are supplemented to form an updated input sample set.

[0067] S54. Using the updated input sample set as input data, re-execute the extreme gradient boosting tree algorithm model construction and training process described in step S2 to obtain a new feature importance score.

[0068] S55. Based on the new feature importance score, re-execute the recursive feature elimination process described in step S3 to output a new optimal feature subset;

[0069] S56. The updated input sample set and the new optimal feature subset are used as the basis for the heart failure risk scoring process in the next cycle, so as to realize the dynamic optimization and adaptive iteration of the risk scoring mechanism.

[0070] The beneficial effects of this invention are:

[0071] This invention deeply integrates the extreme gradient boosting tree algorithm with the recursive feature elimination method to construct an intelligent early warning method suitable for remote monitoring of heart failure patients, significantly improving the accuracy of heart failure risk assessment and the timeliness of system response. Compared with the problems of existing technologies, such as the lack of dynamic optimization in feature selection and the susceptibility of the model to interference from redundant features, this invention introduces an iterative elimination mechanism based on feature importance scoring at the feature level, effectively selecting the most valuable feature subset for risk identification, reducing model complexity and enhancing generalization ability. Simultaneously, the model training process further improves model stability and training efficiency by introducing a third-order derivative acceleration convergence mechanism, adaptive learning rate control, graph structure feature embedding, and structure regularization optimization.

[0072] This invention establishes a multi-level risk threshold classification and feedback comparison mechanism, which not only provides precise early warnings for high-risk events but also records actual processing results for model updates, constructing a dynamic closed loop between risk scores and actual clinical feedback. Through continuous accumulation of feedback samples and label updates, the model can continuously adapt and adjust during operation, possessing robustness and evolution capabilities for long-term deployment scenarios. The system as a whole integrates functional modules such as data acquisition, feature engineering, risk scoring, early warning response, and model retraining, forming a complete link covering "perception-judgment-response-optimization," suitable for continuous monitoring and proactive intervention in remote health management scenarios.

[0073] This invention, while ensuring the accuracy of heart failure risk identification, significantly improves the interpretability, adaptability, and system practicality of the model, and has good medical auxiliary value and prospects for widespread application. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1This is a schematic diagram of the structure of the Internet of Things-based remote monitoring and early warning system for heart failure patients proposed in this invention;

[0076] Figure 2 This is a flowchart illustrating the method for remote monitoring and early warning of heart failure patients based on the Internet of Things proposed in this invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0078] refer to Figure 1 The Internet of Things-based remote monitoring and early warning system for heart failure patients includes the following modules:

[0079] The data acquisition and preprocessing module is used to collect multi-source physiological and behavioral data and preprocess them to generate an input sample set.

[0080] The feature evaluation module is used to build and train an extreme gradient boosting tree algorithm model and obtain feature importance scores.

[0081] The feature selection module is used to execute the recursive feature elimination process and output the optimal feature subset;

[0082] The risk assessment module is used to input the optimal feature subset into the extreme gradient boosting tree algorithm model to generate a heart failure risk score.

[0083] The early warning triggering module is used to trigger an early warning and send warning information to the patient and the platform based on the comparison between the heart failure risk score and the preset risk threshold.

[0084] The feedback recording module is used to record the warning score, warning level and platform response results, and generate feedback information;

[0085] The dynamic optimization module is used to update the input sample set based on feedback information, periodically retrain the extreme gradient boosting tree algorithm model and select features.

[0086] refer to Figure 2 A method for remote monitoring and early warning of heart failure patients based on the Internet of Things includes the following steps:

[0087] S1. Collect multi-source physiological and behavioral data of heart failure patients through IoT devices, preprocess the multi-source physiological and behavioral data, and construct a structured input sample set;

[0088] S2. Construct an extreme gradient boosting tree algorithm model, train it using the input sample set, and obtain the feature importance score corresponding to each input feature;

[0089] S3. Based on the feature importance score, execute the recursive feature elimination process. In each iteration, remove the feature with the lowest score and retrain the extreme gradient boosting tree algorithm model until the set optimal feature subset size or performance convergence condition is met, and output the optimal feature subset.

[0090] S4. Based on the optimal feature subset, the input sample set is discriminated and the heart failure risk score within the corresponding time window is output. Multi-level risk thresholds are set according to the heart failure risk score. When any heart failure risk score exceeds the set risk threshold, the early warning process is triggered, an early warning signal is output and an early warning information is sent to the patient terminal device and the medical platform. At the same time, the early warning event and the processing result are recorded as feedback information.

[0091] S5. Compare the feedback information with the heart failure risk score, update the input sample set, and periodically re-execute the feature evaluation and screening process described in steps S2 and S3 to form a dynamically optimized risk scoring mechanism.

[0092] This invention provides a remote monitoring and early warning method for heart failure patients based on the fusion of extreme gradient boosting trees and recursive feature elimination, which has significant beneficial effects. This method collects multi-source physiological and behavioral data from patients through IoT devices and constructs a structured input sample set, enabling efficient management and organization of complex, high-dimensional health data. The extreme gradient boosting tree algorithm is used to score the importance of each feature, combined with an iterative screening process using recursive feature elimination to effectively remove redundant and low-contribution features, improving the model's discriminative ability and operational efficiency. The constructed heart failure risk scoring mechanism has a tiered early warning function, automatically triggering corresponding early warning responses based on different risk levels, improving the timeliness of remote monitoring and proactive intervention capabilities. By introducing a comparison mechanism between feedback information and scoring results, continuous updating and optimization of the model's input samples can be achieved, supporting long-term adaptation and dynamic learning to changes in patient status, enhancing the model's robustness and evolutionary capabilities. The overall method realizes a closed-loop process from data collection, feature optimization, risk discrimination to dynamic feedback, possessing advantages such as strong interpretability, high prediction accuracy, fast early warning response, and sustainable system optimization, making it suitable for remote health monitoring scenarios in chronic disease management.

[0093] In this embodiment, the multi-source physiological and behavioral data specifically includes individual vital signs and behavioral parameters data collected through Internet of Things (IoT) devices, such as heart rate, blood pressure, respiratory rate, blood oxygen saturation, body temperature, weight, heart rate variability, sleep state, and body movement level.

[0094] In this embodiment, the preprocessing of multi-source physiological and behavioral data specifically includes data cleaning, missing value imputation, normalization, sliding window segmentation, and feature format unification. Specifically, data cleaning, missing value imputation, normalization, sliding window segmentation, and feature format unification refer to: performing data cleaning on multi-source physiological and behavioral data collected through IoT devices to remove data entries containing outliers, duplicate records, or format errors; then performing missing value imputation, appropriately supplementing missing feature items based on time series context or statistical methods; next, performing normalization to convert feature data of different dimensions to a unified numerical range to eliminate the impact of dimensional differences on model training; subsequently, performing sliding window segmentation, dividing the original time series into multiple sample segments according to the set time window length and step size, with each segment serving as an independent input sample; finally, performing feature format unification, aligning and resampling data from different devices and frequencies to unify them into a feature representation format with consistent dimensions and structure, forming a standardized input sample set.

[0095] In this embodiment, S2 specifically includes:

[0096] S21, The input sample set is ,in Indicates the first The original feature vector of each sample, This indicates the corresponding heart failure risk label. The number of samples;

[0097] S22. Construct a graph structure relationship diagram among input features. , where the set of nodes Represents input features, edge set The edges represent the correlation between features, and the edge weights are calculated based on the Pearson correlation coefficient between the features.

[0098] S23. Utilize a graph convolutional network to perform graph structure embedding processing on each sample in the structured input sample set, generating graph structure embedding vectors that fuse feature dependencies. The graph structure embedding process refers to taking the original feature vector of each sample in the structured input sample set, combining it with the pre-constructed feature graph structure relationship graph, and using a graph convolutional network to perform graph convolution operations on the feature nodes, so as to fully integrate the structural dependencies between each feature and its neighboring features and generate a graph structure embedding vector containing global structural information.

[0099] S24. Concatenate the graph structure embedding vector with the original feature vector to form the enhanced input vector. And construct an enhanced input sample set. ;

[0100] S25. Based on the enhanced input sample set, an extreme gradient boosting tree algorithm model is constructed. The extreme gradient boosting tree algorithm model consists of multiple regression trees. The enhanced input vector is used as the input of the extreme gradient boosting tree algorithm model. Through iterative training, the output of each regression tree is continuously accumulated to gradually approach the heart failure risk label value of the corresponding sample.

[0101] S26. Define a third-order approximation objective function based on the enhanced input vector. :

[0102] ;

[0103] in, , , These represent the first, second, and third derivatives of the loss function with respect to the model's predicted values, respectively. This represents the structure regularization term in the objective function. Let t be the augmented input vector of the regression tree for the i-th sample. The output value, This represents the function of the t-th regression tree generated in the t-th round of training;

[0104] objective function The practical significance lies in improving the convergence efficiency and prediction accuracy of the extreme gradient boosting tree algorithm in heart failure risk prediction tasks. This objective function not only incorporates first and second derivative information to capture the direction and curvature of prediction errors, but also further introduces a third derivative, thereby enhancing the model's understanding of the objective function's shape, optimizing the construction direction and step size of each regression tree, and improving the precision and convergence speed of gradient updates. In medical tasks, especially in highly sensitive human health monitoring scenarios, even slight changes in prediction errors can lead to serious consequences. Therefore, introducing a third derivative to improve the objective function's responsiveness to small error fluctuations has significant practical value. The objective function incorporates a structure regularization term to constrain the tree structure complexity and prevent overfitting, thus enhancing the model's expressive power while ensuring generalization performance. Overall, this third-order objective function, while maintaining the high efficiency advantages of traditional extreme gradient boosting trees, significantly improves the model's fine-grained modeling capabilities, making it more suitable for clinical decision-making tasks such as heart failure risk scoring.

[0105] S27. Constructing the structure regularization term in the objective function :

[0106] ;

[0107] in, Indicates the first The number of leaf nodes in the regression tree. For the first The prediction weights of each leaf node, This represents the information entropy value calculated based on the number of samples at each leaf node. , , The structural regularization coefficient;

[0108] Structural regularization terms The practical significance of this regularization term lies in effectively constraining the complexity of the extreme gradient boosting tree algorithm model, preventing overfitting during training, and thus improving its generalization ability and stability in the heart failure risk prediction task. This regularization term comprehensively considers three factors: First, by limiting the number of leaf nodes in the tree, it controls the structural scale of the model, avoiding excessive model complexity expansion caused by excessive splitting; second, it introduces a squared penalty for the prediction weights of leaf nodes to prevent excessive fluctuations in prediction results from a single node; third, it uses an information entropy term based on sample distribution to measure the balance of leaf node partitioning, guiding the model to generate a more reasonable structure during the splitting process, thereby enhancing the model's robustness and interpretability. In scenarios for remote monitoring and early warning of heart failure patients, the model not only requires accurate predictions but also good stability and tolerance to boundary samples. Therefore, this structural regularization term plays a crucial balancing role in the objective function, ensuring that the model achieves the optimal trade-off between performance and complexity, adapting to the dual requirements of reliability and real-time performance in actual medical early warning systems.

[0109] S28. Set the initial predicted values ​​of all samples as a constant term. And set the initial learning rate to ;

[0110] S29, in the During the training rounds, the current model residual descent rate is taken into account. Cumulative forecast error volatility and the dynamic range width of the previous round of heart failure risk score results. Dynamically calculate adaptive learning rate Adaptive control of the update step size during extreme gradient boosting tree algorithm model training:

[0111] ;

[0112] in, The initial learning rate, , , As a regulatory factor;

[0113] Adaptive learning rate Based on three core factors—the rate of residual descent, the volatility of cumulative prediction error, and the dynamic range width of the previous round of heart failure risk scoring results—the model's performance changes in each training round are evaluated in real time, and the update step size is dynamically adjusted accordingly. Specifically, when the model exhibits a stable error descent trend and small prediction fluctuations, the learning rate is appropriately increased to accelerate convergence; conversely, when error fluctuations are large or predictions are unstable, the learning rate is decreased to avoid oscillations and overfitting. This mechanism significantly differs from traditional fixed learning rate strategies, allowing for flexible adjustment of the optimization step based on the characteristics of different stages during training. In IoT-based remote monitoring and early warning scenarios for heart failure patients, data distribution exhibits individual differences and temporal variability. This adaptive strategy ensures that the model can continuously adapt to the learning needs under different data contexts, thereby enhancing the accuracy and robustness of risk scoring results and improving the overall system's intelligent response capability.

[0114] S210. Based on the objective function and the structural regularization term, a greedy splitting strategy is used to construct a regression tree in the current round. and with adaptive learning rate Update the prediction output of the current extreme gradient boosting tree algorithm model;

[0115] S211. Repeat steps S25 to S210 until the extreme gradient boosting tree algorithm model reaches the set upper limit of the number of regression trees or the objective function satisfies the convergence condition.

[0116] S212. Based on the gain contribution of each input feature to the objective function in the trained extreme gradient boosting tree algorithm model, calculate the feature importance score for each input feature. Output feature importance score set , where M is the feature dimension.

[0117] This invention structurally improves the extreme gradient boosting tree algorithm model by introducing a graph structure embedding mechanism and a multi-derivative optimization strategy, significantly enhancing the model's expressive power and learning efficiency in the heart failure risk prediction task. It utilizes graph convolutional networks to extract structural correlation information between input features, forming graph structure embedding vectors that fuse feature relationships. This effectively overcomes the limitations of the traditional feature independence assumption and improves the modeling ability for complex feature interactions. By concatenating the graph structure embedding vectors with the original features, an enhanced input sample set is constructed, improving the information density and discriminative ability of the model input. Introducing a third derivative to construct an approximate objective function more accurately captures gradient change trends and accelerates model convergence; the constructed structure regularization term effectively suppresses overfitting risk and enhances model generalization performance. During model training, the learning rate is dynamically adjusted through joint judgment of the residual descent rate, prediction error volatility, and the dynamic range of the risk score, achieving adaptive control of the model training step size. Based on the feature's contribution to the gain of the objective function, a feature importance score is output, providing efficient support for subsequent feature selection and risk discrimination. The overall approach combines the advantages of graph neural network modeling with the strong generalization ability of gradient boosting trees, resulting in the benefits of accurate modeling, efficient training, and controllable structure.

[0118] In this embodiment, S3 specifically includes:

[0119] S31, Receive input sample set ,in Indicates the first The original feature vector of each sample, This indicates the corresponding heart failure risk label. The number of samples;

[0120] S32. Based on the set of feature importance scores output. Introducing a feature distribution consistency weight coefficient Calculate the weighted score vector The feature distribution consistency weight coefficient is obtained by calculating and normalizing the distribution differences of features in each sub-sample set.

[0121] ;

[0122] Where M is the feature dimension;

[0123] S33, Based on weighted scoring vector Sort in ascending order and select the one with the lowest score. 1 feature, construct a feature index set to be removed. From the input sample set The features corresponding to the feature index set to be removed are removed from the sample set to construct the first round of feature subset. ;

[0124] S34. The first round of feature subset sample set The input is fed into an extreme gradient boosting tree algorithm model for training, and the corresponding feature importance score vector is output. ;

[0125] S35. Set of feature indices to be removed Each feature in Calculate the marginal impact on the objective function of the extreme gradient boosting tree algorithm model. The calculation of the marginal impact on the objective function of the extreme gradient boosting tree algorithm model refers to, in the current round of feature elimination iteration, for each feature in the feature index set to be removed, constructing input sample sets before and after removing the feature, and inputting them into the extreme gradient boosting tree algorithm model for training, recording the minimum value of the corresponding objective function. By comparing the difference in the objective function value before and after removing the feature, the marginal gain or loss of the feature on the overall model loss is calculated, thereby measuring the actual impact strength of the feature on the model performance.

[0126] S36. Let the current iteration number be... Based on the previous round of feature subset sample set and feature importance score vector Perform the following steps:

[0127] Calculate the feature retention score based on the marginal impact. :

[0128] ;

[0129] in, For feature fusion weights, For the first The feature importance score of the j-th feature calculated in the extreme gradient boosting tree algorithm model during round iteration;

[0130] Feature Preservation Score The practical significance lies in quantifying the retention value of each feature for model performance in the current iteration, thus providing a scientific and controllable basis for removal during the recursive feature elimination process. A "feature retention score" is constructed by integrating two key factors: the feature importance score in the extreme gradient boosting tree algorithm model during the previous training round and its marginal impact on the model's objective function. The feature importance score reflects the direct contribution of the feature to the model's predictive performance, while the marginal impact measures the sensitivity of the objective function to changes after removing the feature. By introducing a fusion weight λ, the relative weights of the two in feature evaluation can be flexibly controlled, adapting to the model optimization needs of different scenarios. This scoring mechanism helps avoid the loss of important information due to over-reliance on a single evaluation indicator, and more accurately distinguishes between "redundant features" and "key features" during the feature removal stage. Especially in remote monitoring and early warning systems for heart failure patients, where feature sources are wide-ranging and highly heterogeneous, this provides a multi-dimensional and interpretable feature retention evaluation method, offering a reliable basis for building a concise and efficient model, and effectively improving the stability and accuracy of the final risk scoring results.

[0131] By combining feature retention scores and feature distribution consistency weighting coefficients, a tiered elimination strategy is implemented to construct a new set of feature subset samples. ;

[0132] The new round of feature subset sample set The input is fed into an extreme gradient boosting tree algorithm model for training, and the corresponding feature importance score vector is output. ;

[0133] Calculate the performance metrics of the extreme gradient boosting tree algorithm model in this round. The calculation of the performance metrics for the extreme gradient boosting tree algorithm model in this round is based on the model's prediction performance on the validation set or cross-validation subset after the current round of training. Commonly used evaluation metrics include mean absolute error, mean squared error, precision, recall, or F1 score. Specifically, the trained model is applied to the validation data to obtain the error between the model's predicted output and the true labels. The performance value for the current round is then calculated using the selected metric formula. Simultaneously, the performance metrics can be compared with those of the previous round to determine whether the model's performance has improved, whether it has reached the convergence criterion, or whether it meets the performance degradation tolerance threshold, thus assisting in determining whether the iteration process should continue.

[0134] S37. Determine if the iteration termination condition is met. If it is, terminate the recursive feature elimination process; otherwise, let... Return to step S36 and continue execution. The iteration termination conditions include:

[0135] Input dimension in the current feature subset sample set Less than the set minimum feature dimension ;

[0136] The performance degradation of the extreme gradient boosting tree algorithm model meets the following requirements: ,in This is the performance tolerance threshold;

[0137] S38. Denote the iteration termination round as... Extract the currently retained set of feature indices. , as the final optimal feature subset;

[0138] S39. Based on the feature index set For structured input sample sets Perform feature mapping on all samples to generate the final input sample set. .

[0139] This invention introduces a feature distribution consistency assessment and marginal impact analysis mechanism into the recursive feature elimination process, significantly improving the scientific rigor of feature selection and the stability of model performance. By constructing a feature distribution consistency weight coefficient, the distribution stability of features across different subsets is measured, effectively avoiding fluctuations in importance scores caused by sample perturbations and enhancing the robustness of feature selection. A weighted scoring vector is introduced, combining the dual dimensions of feature importance and distribution consistency, ensuring that both predictive contribution and distribution stability are considered during feature removal. Furthermore, by calculating the marginal impact of candidate features to be removed on the objective function, the actual impact on model performance after removal is evaluated, improving the accuracy of feature removal decisions. During iteration, a hierarchical removal strategy is executed based on feature retention scores and distribution consistency weights, gradually constructing a high-quality feature subset. The iteration termination condition is dynamically determined by integrating model performance evaluation results, ensuring no performance loss during feature compression. The final optimal feature subset output maintains the model's generalization ability while achieving effective compression of the input dimension, reducing system computational overhead and improving the model's real-time performance and deployment efficiency. The overall process integrates stability evaluation and performance feedback adjustment, possessing higher interpretability and adaptability, making it particularly suitable for risk screening scenarios based on high-dimensional health monitoring data.

[0140] In this embodiment, S4 specifically includes:

[0141] S41. Receive the final input sample set. ,in, Indicates the first The input feature vector of each sample under the selected optimal feature subset dimension. This indicates the corresponding heart failure risk label. The number of samples;

[0142] S42. Calculate the optimal feature vector of each sample in the final input sample set. The input is fed into the constructed and trained extreme gradient boosting tree algorithm model to perform the classification and discrimination process for heart failure risk, and outputs the heart failure risk score for each sample within a set time window. The heart failure risk score is formed by inputting the optimal feature subset vector of each sample in the final input sample set into the trained extreme gradient boosting tree algorithm model. The sample is jointly predicted by multiple decision regression trees built inside the model. The output results of each regression tree are accumulated to generate the heart failure risk score of the sample in the current time window. The heart failure risk score is used as the quantitative judgment result of the degree of heart failure risk.

[0143] S43. Construct a risk score vector from the heart failure risk scores corresponding to all samples. ,in Indicates the first The heart failure risk score results of each sample in the classification and discrimination process;

[0144] S44. Set a multi-level risk threshold set Among them, satisfying , Indicates the first Each risk level corresponds to a lower limit threshold for scoring, and a total of [number] thresholds are constructed. One risk level range;

[0145] S45. For any score in the risk scoring vector Determine the risk threshold set within the established multi-level risk threshold range. Within the risk level range to which it belongs, there is a unique risk level label. Satisfying the inequality and label the risk level As the first Output of heart failure risk level for each sample;

[0146] S46. Assemble a risk level label vector from all the risk level labels corresponding to the samples. ,in Represents the input feature vector The corresponding heart failure risk level result;

[0147] S47. Set the risk level threshold for triggering early warnings. When any risk level label If this occurs, immediately trigger the early warning process and initiate the heart failure risk emergency response procedure;

[0148] S48. Form an early warning trigger sample set by combining the sample indices that meet the early warning conditions. And targeting the early warning trigger sample set For each sample, the corresponding patient terminal device and the medical service platform send a warning signal and a warning information packet containing a risk level label, a risk score value, and time window information;

[0149] S49. Record the sample number corresponding to each warning event , heart failure risk score value , risk level label , the patient terminal device number sent and the warning response processing information returned by the medical platform, and construct feedback information .

[0150] Through the construction of an extreme gradient boosting tree classification structure based on the optimal feature subset, the present invention realizes the refined discrimination and hierarchical evaluation of heart failure risks, significantly enhancing the accuracy and practicality of the warning system. Using high-quality feature inputs to improve the model's discrimination ability for heart failure risk scores effectively improves the accuracy and stability of the heart failure risk prediction results; setting multiple risk thresholds and outputting risk level labels realizes the interval division and labeling of heart failure risk levels, facilitating medical staff to quickly grasp the patient's risk status. Setting a warning trigger mechanism and response logic, when the risk score exceeds the set threshold, high-risk individuals can be automatically identified and the warning program can be initiated, realizing efficient risk response and multi-terminal linked warning information push, and enhancing the system's emergency response speed and coverage breadth for sudden heart failure events. Establishing a complete warning information recording mechanism ensures that the key data, response actions, and processing results of each warning event are retained by the system, providing detailed data support for evaluation, optimization, and clinical decision-making. The overall technical solution constructs a closed-loop risk discrimination - hierarchical evaluation - warning trigger - information push - response feedback chain, strengthening the system's real-time monitoring and auxiliary diagnosis and treatment capabilities while ensuring prediction performance, and is applicable to the intelligent remote monitoring and risk intervention scenarios of a large number of heart failure patients based on the Internet of Things environment.

[0151] In this embodiment, the S5 specifically includes:

[0152] S51. Extract the sample index, heart failure risk score result, risk level label, and actual response processing result of historical warning events from the feedback information to construct a feedback sample set;

[0153] S52. Compare the scoring results in the feedback sample set with the corresponding feedback information, analyze the accuracy of the extreme gradient boosting tree algorithm model's scoring and the rationality of the risk level classification, and identify samples and their characteristic attributes where the scoring deviation is greater than the preset deviation threshold. Specifically, comparing the scoring results in the feedback sample set with the corresponding feedback information and analyzing the accuracy of the extreme gradient boosting tree algorithm model's scoring and the rationality of the risk level classification means comparing the heart failure risk score value generated in the model for each historical warning sample and its corresponding risk level label with the actual clinical response results one by one to determine whether the model score accurately reflects the severity of the condition and whether the risk level classification is reasonable and triggers an effective warning, thereby identifying samples with obvious deviations and analyzing their corresponding characteristic attributes and label setting problems.

[0154] S53. Based on the comparison analysis results, the relevant sample labels in the original input sample set are corrected, and the newly added feedback samples and their latest label information are supplemented to form an updated input sample set.

[0155] S54. Using the updated input sample set as input data, re-execute the extreme gradient boosting tree algorithm model construction and training process described in step S2 to obtain a new feature importance score.

[0156] S55. Based on the new feature importance score, re-execute the recursive feature elimination process described in step S3 to output a new optimal feature subset;

[0157] S56. The updated input sample set and the new optimal feature subset are used as the basis for the heart failure risk scoring process in the next cycle, so as to realize the dynamic optimization and adaptive iteration of the risk scoring mechanism.

[0158] This invention constructs a feedback sample set by introducing feedback information, achieving adaptive updating and dynamic optimization of the scoring mechanism. By extracting and analyzing the risk scores and response processing results of historical warning events, samples with large scoring errors and key features can be accurately identified, thereby timely correcting the input sample set and label data, supplementing with new feedback samples, and enhancing the timeliness and representativeness of the sample set. Retraining the extreme gradient boosting tree algorithm model and updating the feature importance score helps to construct an optimal feature subset that better reflects current clinical practice. This not only enhances the model's adaptability to new feature patterns but also effectively improves the accuracy of risk scoring and the credibility of warning judgments. By constructing a closed-loop optimization mechanism, the heart failure risk identification system can dynamically respond to real feedback, achieving model self-correction and performance iterative updates, significantly improving the system's intelligence level and the effectiveness of remote health management.

[0159] Example 1:

[0160] To verify the feasibility of this invention in practice, it was applied to the cardiovascular chronic disease management center of a tertiary hospital. To improve the quality of follow-up care for patients with chronic heart failure after discharge, the research team selected 10 typical heart failure patients who received treatment and were discharged between May 10th and May 24th, 2025, to conduct a pilot study on remote monitoring based on this invention. The patients had an average age of 64 years and all suffered from grade 2 or higher left ventricular dysfunction, accompanied by typical symptoms such as paroxysmal nocturnal dyspnea and decreased exercise tolerance.

[0161] Before discharge, all patients were equipped with wearable devices with IoT communication modules, including a continuous ECG chest patch, a wireless blood pressure monitor, a sleep apnea wristband, and a Bluetooth pulse oximeter. The devices are connected to the hospital's information platform via home Wi-Fi and upload seven types of multi-source physiological and behavioral data daily, including heart rate, blood pressure, respiratory rate, blood oxygen saturation, frequency of nighttime awakenings, and activity level.

[0162] The platform's backend first calls the data preprocessing module to perform standardization, noise removal, and time window alignment, constructing a structured input sample set. This sample set is then fed into an improved extreme gradient boosting tree model, which integrates graph structure feature embedding and a third-order derivative to accelerate the objective function, enhancing its ability to capture nonlinear risk factors. After the model outputs an importance score for each feature, the recursive feature elimination module initiates a hierarchical elimination and marginal impact calculation process, generating a 10-dimensional optimal feature subset while ensuring model performance stability.

[0163] Based on this, the model performs risk scoring on the samples uploaded each day, setting risk thresholds at four levels: normal (<0.5), attention (0.5–0.7), high risk (0.7–0.85), and very high risk (>0.85). Once the score exceeds 0.85, the platform immediately sends an alert to the patient's app and the medical staff workstation, and records the event number, alert level, timestamp, and subsequent processing results.

[0164] Table 1. Comparison of the effects of traditional methods and the method of this invention

[0165]

[0166] The data in Table 1 clearly demonstrates that the method of this invention has superior performance compared to traditional methods in remote monitoring and early warning of heart failure patients. In the traditional method, only four patients (P003, P006, P008, and P010) were successfully alerted, and their corresponding risk scores were all high. The other six patients, even those exhibiting certain risk characteristics, were not identified in a timely manner, potentially leading to missed reports and affecting the timeliness of intervention.

[0167] In contrast, the method of this invention combines extreme gradient boosting trees with recursive feature elimination techniques. After fully exploring the complex correlations between input features, it successfully and accurately identified all high-risk patients. A total of 8 patients were triggered with early warnings (including patients such as P001, P002, P005, P007, and P009, who were not alerted by traditional methods), and all early warning results were consistent with the actual clinical intervention results. Taking P001 as an example, the traditional method scored 0.58 and did not trigger an early warning, while the method of this invention identified it as 0.76 and initiated remote diagnosis, successfully intervening in a potential emergency. Similarly, for P007 and P009, traditional methods did not issue early warnings for scores below 0.5, but the method of this invention identified them as 0.65 and 0.63 respectively, and ultimately, doctors contacted them for follow-up and intervention via a remote system.

[0168] It is worth noting that the method of this invention did not produce any false alarms. For example, patient P004 did not trigger any warnings under either method, and no changes in risk were reported clinically, further demonstrating the system's strong stability and robustness. Furthermore, patients P006, P008, and P010 already had high scores, but the present invention further improved their scores, effectively amplifying the danger signals and ensuring the foresight and urgency of the medical response.

[0169] In summary, the results fully verify that the method of the present invention is not only significantly superior to traditional methods in terms of risk scoring sensitivity, but also has higher consistency in clinical intervention. It has important practical significance for proactive prevention and control, personalized management and rational allocation of medical resources for heart failure patients in the Internet of Things environment.

[0170] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A remote monitoring and early warning system for heart failure patients based on Internet of Things, characterized in that, Comprise the following modules: Data acquisition and preprocessing module, for collecting multi-source physiological and behavioral data and preprocessing, generating input sample set; Feature evaluation module, for constructing and training extreme gradient boosting tree algorithm model, obtaining feature importance score; Feature selection module, for executing recursive feature elimination process, outputting optimal feature subset; Risk discrimination module, for inputting the optimal feature subset into the extreme gradient boosting tree algorithm model, generating heart failure risk score; Early warning trigger module, for comparing heart failure risk score with preset risk threshold, triggering early warning and sending early warning information to patients and platform; Feedback recording module, for recording early warning score, early warning level and platform response result, generating feedback information; Dynamic optimization module, for updating input sample set based on feedback information, periodically retraining extreme gradient boosting tree algorithm model and screening features; The feature evaluation module is specifically used for: S21, The input sample set is ,in Indicates the first The original feature vector of each sample, This indicates the corresponding heart failure risk label. The number of samples; S22, constructing a graph structure relationship graph between input features wherein a node set represents input features, and an edge set represents correlation edges between features, and an edge weight is obtained according to a Pearson correlation coefficient between features; S23, performing graph structure embedding processing on each sample in the structured input sample set by using a graph convolution network to generate a graph structure embedding vector fusing feature dependency relationship ; S24, concatenate the graph structure embedding vector with the original feature vector to form an enhanced input vector , and construct an enhanced input sample set ; S25, based on the enhanced input sample set, constructing an extreme gradient boosting tree algorithm model, the extreme gradient boosting tree algorithm model is composed of multiple regression trees, the enhanced input vector is used as the input of the extreme gradient boosting tree algorithm model, the output of each regression tree is continuously accumulated through iterative training, and the heart failure risk label value of the corresponding sample is gradually approached; S26, defining a third order approximation of the target function based on the enhanced input vector ; ； wherein, , , respectively represent the first, second and third order derivatives of the loss function with respect to the model prediction, represents the structured regularizer in the objective function, is the output value of the t-th regression tree for the i-th sample enhanced input vector , represents the t-th regression tree function generated in the t-th round of training; S27、constructing a structure regular term in the objective function ; ； wherein, denotes the number of leaf nodes of the is the prediction weight of the denotes the information entropy value calculated based on the number of samples of each leaf node, , , is the structural regularity coefficient;​​ S28, set the initial prediction value of all samples as a constant term and set the initial learning rate as ; S29, in the first In the training of the wheel, considering the current model residual decline rate , cumulative prediction error fluctuation And the dynamic interval width of the last round of heart failure risk score results , dynamically calculate the adaptive learning rate , adaptive control update step in the process of extreme gradient boosting tree algorithm model training; ； wherein, is an initial learning rate, , , is a regulatory factor; S210, based on the target function and the structure regularization term, a regression tree is constructed in the current round by using a greedy splitting strategy , and an adaptive learning rate updating the prediction output of the current extreme gradient boosting tree algorithm model; S211, repeat steps S25 to S210 until the extreme gradient boosting tree algorithm model reaches the set upper limit of the number of regression trees or the objective function meets the convergence condition; S212、based on the gain contribution degree of each input feature to the target function in the trained extreme gradient boosting tree algorithm model, calculate the feature importance score of each input feature , output the set of feature importance scores , where M is the feature dimension The feature selection module is specifically used for: S31, receiving an input sample set wherein represents an original feature vector of an i-th sample, represents a corresponding heart failure risk label, is a sample number;​ S32、according to the output feature importance score set , introducing a feature distribution consistency weight coefficient , calculating a weighted score vector wherein the feature distribution consistency weight coefficient is obtained by calculating the distribution difference degree of the feature in each sub-sample set and normalizing it: ； Wherein, M is the feature dimension; S33、based on the weighted score vector performing ascending order sorting, obtaining the feature with the lowest score , constructing a feature index set to be removed , removing the features corresponding to the feature index set to be removed from the input sample set , constructing a first round feature subset sample set ; S34, the first round of feature subset sample set input into the extreme gradient boosting tree algorithm model for training, output the corresponding feature importance score vector ; S35、to the feature index set to be removed each feature in the feature index set , calculate the marginal influence quantity of the objective function of the extreme gradient boosting tree algorithm model ; S36, let the current iteration round be based on the previous round feature subset sample set and the feature importance score vector perform the following steps: According to the marginal impact amount, a feature retention score is calculated ; ； wherein, is a feature fusion weight, is a feature importance score of the jth feature in the tth iteration calculated in the extreme gradient boosting tree algorithm model. In combination with the feature retention score and the feature distribution consistency weight coefficient, a stratification elimination strategy is executed to construct a new round of feature subset sample set ; a new round of feature subset sample set input into an extreme gradient boosting tree algorithm model for training, and output a corresponding feature importance score vector ; Computing performance indicators for the current round of extreme gradient boosting tree algorithm model ; S37, judging whether an iteration termination condition is satisfied, if yes, terminating the recursive characteristic elimination process, otherwise, letting , returning to step S36 for continuous execution, the iteration termination condition comprising: input dimensions in the current feature subset sample set less than a set minimum feature dimension ; The extreme gradient boosting tree algorithm model performance decline meets wherein is a performance tolerance threshold; S38, record the iteration termination round as , extract the current reserved feature index set , as the final optimal feature subset; S39, according to the feature index set performing feature mapping on all samples in the structured input sample set to generate a final input sample set ​ The risk discrimination module and the early warning trigger module are specifically used for: S41. Receive the final input sample set. ,in, Indicates the first The input feature vector of each sample under the selected optimal feature subset dimension. This indicates the corresponding heart failure risk label. The number of samples; S42、obtaining the optimal feature vector of each sample in the final input sample set The extreme gradient boosting tree algorithm model is input into the constructed and trained completed model, a classification and discrimination process of heart failure risk is performed, and a heart failure risk score value of each sample in a set time window is output , to constitute a heart failure risk score result; S43, forming a risk score vector with the heart failure risk score values corresponding to all samples wherein represents the heart failure risk score result of the i-th sample in the classification discrimination process; and the i-th sample in the classification discrimination process. S44, set a multi-level risk threshold set wherein, meet , represent the lower threshold of the score corresponding to the risk level, a total of risk level interval is constructed; S45, for any score in the risk score vector , determining a risk level interval to which the set of multiple risk threshold values belonging to the interval of risk levels, there is only one risk level label satisfying the inequality , and outputting the risk level label as the heart failure risk level of the th sample; S46, forming a risk level label vector by using the risk level labels corresponding to all the samples wherein represents the input feature vector corresponding heart failure risk level result; S47, set the risk level threshold of early warning trigger When any risk level label Immediately trigger the early warning process to start the heart failure risk emergency response program; S48, grouping the samples satisfying the early warning condition into an early warning trigger sample set and sending, for each sample in the early warning trigger sample set an early warning signal and an early warning information package containing a risk level label, a risk score value, and time window information to the corresponding patient terminal device and the medical service platform. S49, record the sample number corresponding to each early warning event , heart failure risk score value , risk level label , the patient terminal device number sent and the early warning response processing information returned by the medical platform to construct feedback information .

2. The IoT based remote monitoring and alerting system for heart failure patients as claimed in claim 1 wherein, Multi-source physiological and behavioral data specifically include individual vital signs and behavioral parameter data of heart rate, blood pressure, respiratory rate, oxygen saturation, body temperature, body weight, heart rate variability, sleep state and body movement level collected through Internet of Things devices.

3. The IoT based remote monitoring and alerting system for heart failure patients as claimed in claim 1 wherein, The preprocessing of multi-source physiological and behavioral data specifically includes data cleaning, missing value filling, normalization processing, sliding window segmentation and feature format unification.

4. The IoT based remote monitoring and alerting system for heart failure patients as claimed in claim 1 wherein, The dynamic optimization module is specifically used for: S51, extracting sample index, heart failure risk score result, risk level label and actual response processing result of historical early warning events from feedback information, constructing feedback sample set; S52, comparing the score result in the feedback sample set with the corresponding feedback information, analyzing the scoring accuracy and risk level division rationality of the extreme gradient boosting tree algorithm model, and identifying the samples and feature attributes with a scoring deviation greater than a preset deviation threshold; S53, according to the comparison and analysis result, correcting the related sample label in the original input sample set, supplementing the new feedback sample and its latest label information, and forming the updated input sample set; S54, taking the updated input sample set as input data, re-executing the extreme gradient boosting tree algorithm model construction and training process in step S2, and obtaining new feature importance score; S55, based on the new feature importance score, re-executing the recursive feature elimination process in step S3, outputting a new optimal feature subset; S56, taking the updated input sample set and the new optimal feature subset as the basis of the next cycle of heart failure risk scoring processing, realizing dynamic optimization and adaptive iteration of the risk scoring mechanism.

Images