CN-121983304-A - Obesity comprehensive evaluation model based on user health data and application thereof

CN121983304ACN 121983304 ACN121983304 ACN 121983304ACN-121983304-A

Abstract

The invention relates to the technical field of intelligent medical treatment and discloses an obesity comprehensive evaluation model based on user health data and application thereof. The method breaks through the limitation of traditional single data evaluation by integrating multi-source heterogeneous health data, can comprehensively capture multi-dimensional factors affecting obesity, combines data preprocessing and characteristic engineering optimization, effectively improves data and characteristic quality, lays a foundation for accurate evaluation, and secondly, ensures feature extraction and evaluation accuracy by adopting a deep learning and causal reasoning mixed architecture, and improves evaluation reliability by definitely analyzing the action logic of features on obesity evaluation through interpretability analysis, and in addition, generates dynamic obesity risk scores updated along with real-time health data of users, can reflect risk changes in real time, and provides accurate and efficient technical support for personalized health management and obesity intervention.

Inventors

GONG GUIFANG
GENG LANLAN
MAO XIAOJIAN
TANG GUOYI
GUO YUHUAI
JIANG HONGLI
LIANG CUIPING
ZHANG XIAOCHUN
ZUO LIANDONG

Assignees

广州医科大学附属妇女儿童医疗中心

Dates

Publication Date: 20260505
Application Date: 20251226

Claims (10)

1. An obesity integrated assessment model based on user health data, comprising: the data acquisition module is used for acquiring multi-source heterogeneous health data of a user, wherein the multi-source heterogeneous health data comprises physiological index data, wearable equipment dynamic monitoring data, diet recording data and environmental parameter data; The data preprocessing module is used for carrying out cleaning, standardization and cross-modal fusion processing on the multi-source heterogeneous health data, extracting multi-dimensional initial characteristics from the processed data and outputting standardized multi-dimensional initial characteristic data; The feature engineering module is used for receiving the output of the data preprocessing module, executing feature screening, feature conversion and feature fusion operation, eliminating low-value features, mapping multi-dimensional data to a unified feature space and outputting an engineering feature set for model training; the model training module is based on a deep learning and causal reasoning mixed architecture and comprises: the basic layer is used for receiving the engineering feature set, respectively extracting the features of different types of data in the engineering feature set through a deep learning network, and calculating and outputting corresponding feature weights based on the preliminary association degree of each feature and obesity evaluation; the optimizing layer is used for receiving the output of the base layer, dynamically adjusting the characteristic weight through a weight optimizing algorithm and outputting the adjusted characteristic weight and corresponding characteristic data; The interpretation layer is used for receiving the output of the optimization layer, analyzing the influence weight of each characteristic on the obesity evaluation result through an interpretability analysis algorithm, generating an interpretable rule set, and feeding the analyzed influence weight back to the optimization layer; And the evaluation output module is used for receiving the output of the model training module and generating a dynamic obesity risk score based on the feature data output by the optimization layer, the influence weight and the interpretable rule set, wherein the dynamic obesity risk score is updated along with the real-time health data of the user.
2. The comprehensive evaluation model for obesity based on user health data according to claim 1, wherein the cleaning, normalizing and cross-modal fusion of the multi-source heterogeneous health data and outputting normalized unified format data comprises: preliminary outlier rejection is carried out on the physiological index data and the wearable equipment dynamic monitoring data through an outlier first-stage filtering rule, regional outlier detection is carried out on diet image data in the diet recording data through a convolution network, and effective original data is reserved; converting the physiological index data and the environmental parameter data into [ -1,1] value intervals through a tan-h normalization algorithm, uniformly mapping the time sequence values of the wearable equipment dynamic monitoring data into [0,1] intervals through a Min-Max normalization algorithm, and eliminating dimension differences; And carrying out space-time alignment on the physiological index data, the wearable equipment dynamic monitoring data, the diet recording data and the environment parameter data based on a data acquisition timestamp, extracting multi-dimensional initial characteristics from the aligned data, wherein the multi-dimensional initial characteristics comprise body fat rate values, hourly exercise intensity, food type codes and time period identifiers, and forming and outputting standardized multi-dimensional initial characteristic data.
3. The comprehensive assessment model of obesity based on user health data of claim 1, wherein the feature engineering module comprises: The feature screening unit is configured to calculate variance contribution degree of each initial feature in the standardized multidimensional initial feature data and correlation significance with the obesity evaluation tag, and reject low-value initial features with low variance contribution degree and low correlation significance to form a screened feature set; the feature conversion unit is configured to respectively execute conversion operation on the time sequence features, the text features and the static numerical features in the filtered feature set to generate converted features trained by the adaptation model; And the feature fusion unit is configured to splice and weight fusion the converted features to generate an engineering feature set.
4. The comprehensive evaluation model for obesity based on user health data according to claim 3, wherein the calculation of the variance contribution degree sequentially comprises the steps of: Detecting and eliminating abnormal characteristic values based on a box diagram rule; and detecting redundancy of the residual features through the pearson correlation coefficient, if the absolute value of the pearson correlation coefficient between the two features is larger than the preset threshold, removing the features with lower relevance to the obesity evaluation tag, and retaining the features with higher relevance.
5. The comprehensive evaluation model for obesity based on user health data according to claim 3, wherein the performing of the conversion operation comprises the steps of: extracting derivative time sequence characteristics from the time sequence characteristics by adopting a fixed time interval sliding window method, wherein the derivative time sequence characteristics comprise average motion intensity and night sleep fragmentation times in a fixed time interval; word segmentation coding is carried out on the text features by adopting a word bag model, and text conversion features of food types and cooking modes are extracted; and mapping the static numerical value characteristics to the [0,1] numerical value interval by adopting a normalization algorithm to generate static conversion characteristics.
6. The comprehensive assessment model for obesity based on user health data according to claim 5, wherein the stitching and weighted fusion sequentially comprises the steps of: Splicing the derivative time sequence feature, the text conversion feature and the static conversion feature into an initial feature matrix; And assigning weights based on the feature type identifiers and the preliminary relevancy, wherein the features in the same type are assigned with sub-weights according to the preliminary relevancy absolute value proportion, and finally an engineering feature set is generated.
7. The comprehensive evaluation model for obesity based on user health data according to claim 6, wherein the obtaining of the preliminary association degree sequentially comprises the following steps: Determining the types of all the features in the engineering feature set, wherein the physiological index features and the environmental parameter features are defined as continuous numerical type features, and the food category features and the period identification features are defined as discrete classification type features; calculating the linear correlation strength between the continuous numerical type characteristic and the obesity evaluation tag through the Pearson correlation coefficient; calculating the category association strength between the discrete classification type features and the obesity evaluation tag through chi-square test; And carrying out normalization processing on the calculation results of the linear association strength and the category association strength to obtain the preliminary association degree of each feature and obesity evaluation.
8. The comprehensive evaluation model for obesity based on user health data according to claim 1, wherein the analyzing the influence weight of each feature on the result of the obesity evaluation by the interpretability analysis algorithm generates an interpretable rule set, comprising the steps of: Calculating the difference of obesity evaluation results output by a model as marginal contribution of the target feature under the feature subset for each feature subset when the feature subset contains the target feature and does not contain the target feature based on a saprolidine additively interpretation value algorithm, and carrying out weighted average on the marginal contribution of the target feature according to the scale and the occurrence probability of uniform distribution of each feature subset to obtain the contribution degree of the target feature to the obesity evaluation result and taking the absolute value of the contribution degree as influence weight; The method comprises the steps of selecting characteristics with top influence weight ranking as key characteristics, counting obesity risk values corresponding to different value intervals of each key characteristic, determining association rules of obesity risk rising when a certain key characteristic value exceeds a specific range, referring to health medical standards and combining the value intervals corresponding to healthy people in characteristic value distribution aiming at physiological index characteristics and exercise intensity characteristics, determining threshold rules of normal ranges of different body fat rates and normal ranges of daily exercise intensity, carrying out combination test on the key characteristics, calculating obesity risk values when the key characteristics are acted after combination and comparing the obesity risk values with the risk values when single characteristics are acted, and determining synergistic influence rules of obesity risk rising under the action of combination characteristics.
9. The comprehensive assessment model of obesity based on user health data according to claim 8, wherein the dynamic obesity risk score is generated by: Extracting each piece of characteristic data output by the optimization layer, multiplying each piece of characteristic data by the corresponding influence weight, and summing all multiplication results to obtain a basic obesity risk score; The basic obesity risk score is adjusted by comparing with an interpretable rule set, if certain key feature data exceeds a specific range defined in an association rule, the basic obesity risk score is adjusted up according to the risk rising amplitude determined by the association rule; After the real-time health data of the user is updated, the updated data is sequentially acquired by the data acquisition module, processed by the data preprocessing module and converted into an engineering feature set by the feature engineering module, and then the engineering feature set is processed by the model training module to obtain influence weights and fed back to the optimization layer; and repeatedly calculating a basic risk score and adjusting the score based on the updated key feature data and the corresponding influence weight to obtain an updated dynamic obesity risk score.
10. Use of an integrated assessment model of obesity based on user health data according to any of claims 1-9 in user obesity risk prediction.

Description

Obesity comprehensive evaluation model based on user health data and application thereof Technical Field The invention relates to the technical field of intelligent medical treatment, in particular to an obesity comprehensive evaluation model based on user health data and application thereof. Background In the current obesity evaluation technology, the core technical bottleneck is concentrated on three dimensions of insufficient multi-source data utilization, difficulty in considering model precision and interpretability, and lack of dynamic adaptability of evaluation results. In the data layer, the existing scheme depends on single static physiological indexes such as body mass index, body fat rate and the like, and severely restricts the input quality of the model. On the model architecture level, the traditional scheme or the deep learning architecture is adopted to pursue the precision but lose the interpretability due to the characteristics of the black box, the influence weight and the association rule of each health feature on the obesity risk cannot be defined, or the traditional statistical model is adopted to ensure the interpretability but difficult to fit the complex multidimensional feature, and the precision is limited. And the evaluation result has low adaptation degree with the actual health requirement and insufficient reliability, so that the application value of the evaluation result in obesity risk prediction is limited, and personalized obesity intervention guidance is difficult to support. Disclosure of Invention The invention aims to provide an obesity comprehensive evaluation model based on user health data and application thereof, so as to solve the technical problems in the background technology. In order to achieve the above purpose, the present invention discloses the following technical solutions: In a first aspect, the invention discloses an obesity comprehensive assessment model based on user health data, comprising: the data acquisition module is used for acquiring multi-source heterogeneous health data of a user, wherein the multi-source heterogeneous health data comprises physiological index data, wearable equipment dynamic monitoring data, diet recording data and environmental parameter data; The data preprocessing module is used for carrying out cleaning, standardization and cross-modal fusion processing on the multi-source heterogeneous health data, extracting multi-dimensional initial characteristics from the processed data and outputting standardized multi-dimensional initial characteristic data; The feature engineering module is used for receiving the output of the data preprocessing module, executing feature screening, feature conversion and feature fusion operation, eliminating low-value features, mapping multi-dimensional data to a unified feature space and outputting an engineering feature set for model training; the model training module is based on a deep learning and causal reasoning mixed architecture and comprises: the basic layer is used for receiving the engineering feature set, respectively extracting the features of different types of data in the engineering feature set through a deep learning network, and calculating and outputting corresponding feature weights based on the preliminary association degree of each feature and obesity evaluation; the optimizing layer is used for receiving the output of the base layer, dynamically adjusting the characteristic weight through a weight optimizing algorithm and outputting the adjusted characteristic weight and corresponding characteristic data; The interpretation layer is used for receiving the output of the optimization layer, analyzing the influence weight of each characteristic on the obesity evaluation result through an interpretability analysis algorithm, generating an interpretable rule set, and feeding the analyzed influence weight back to the optimization layer; And the evaluation output module is used for receiving the output of the model training module and generating a dynamic obesity risk score based on the feature data output by the optimization layer, the influence weight and the interpretable rule set, wherein the dynamic obesity risk score is updated along with the real-time health data of the user. Preferably, the cleaning, normalizing and cross-modal fusion processing are performed on the multi-source heterogeneous health data, and standardized unified format data is output, including: preliminary outlier rejection is carried out on the physiological index data and the wearable equipment dynamic monitoring data through an outlier first-stage filtering rule, regional outlier detection is carried out on diet image data in the diet recording data through a convolution network, and effective original data is reserved; Converting the physiological index data and the environmental parameter data into [ -1,1] value intervals through a tan-h normalization algorithm, uniformly mapping the time sequence