A sports risk assessment method and system based on wearable devices

Abstract

The present invention discloses a method and system for sports risk assessment based on wearable devices. It includes: obtaining an initial data set through authorization to train a model, using wearable devices to collect physiological indicator information, optimizing relevant features, and then performing sports risk assessment through a server to generate sports suggestions. User feedback is used to iteratively update the training model. The present invention realizes wearable device data collection, integration, heterogeneous data fusion and sports risk assessment; through the change curve of the collected long-term physiological characteristics, combined with massive user data to perform correlation analysis, binning discrete coding and other data processing, completes the filtering, transformation, optimization and fusion of multi-source data, and extracts more effective data features; finally, the sports risk is assessed by a variety of comprehensive features after data fusion. A sports risk assessment method with high interpretability is realized, providing users with certain suggestions for dealing with sports risks.

Description

Technical Field

[0001] The present invention relates to the fields of smart home and smart health, and in particular to a sports risk assessment method and system based on wearable devices. Background Art

[0002] With rising living standards, people are pursuing healthier lifestyles and are willing to spend more time on exercise. Active and scientific exercise is increasingly becoming a conscious choice. However, in previous exercise and fitness programs, people often faced challenges such as whether the exercise environment was appropriate, whether their physical indicators were healthy, and whether the amount of exercise was appropriate. A single mistake could lead to sports injuries or accidents. Therefore, obtaining appropriate exercise risk assessment advice before exercise is crucial. With the increasing integration of wearable devices and exercise, a growing number of smart exercise applications and devices have emerged to meet people's diverse needs. Smart wearable devices targeting the need for scientific exercise management are also gradually emerging, providing users with a variety of refined services such as exercise planning and exercise risk assessment. Currently, a large number of wearable devices are available on the market, helping people monitor their heart rate, calories burned, and daily steps. Existing wearable-based exercise risk assessment methods often rely heavily on pre-defined rules or simple linear regression, relying heavily on prior knowledge. These methods lack the integration and knowledge discovery of heterogeneous data collected by different wearable devices, making it difficult to achieve scientific and accurate exercise risk assessment and management. The increasing demand for more scientific and convenient exercise has spurred the proliferation of smart wearable devices, providing a foundation for online intelligent big data analysis. This paper proposes deploying a sports risk assessment model based on wearable device data on the server side, conducting big data analysis and mining, and providing users with scientific sports risk assessment results, helping them improve the efficiency and accuracy of sports risk assessment and management.

[0003] One of the current existing technologies is the multifunctional smart wearable devices on the market, such as sports watches and sports bracelets. The existing technical solution: First, by using smart wearable devices to collect different sports information, such as sedentary time, number of steps, etc.; second, the user's physical condition is analyzed through rules and corresponding exercise recommendations are given. The shortcomings of the existing technical solution: First, this type of product is a highly versatile wearable device, so the collected data is not targeted for the task of sports risk assessment; second, the product lacks analysis services that combine massive data, the scope of application is limited, and the scalability is not high.

[0004] The second existing technology is the motion detection equipment currently available in fitness centers. This existing technology solution: First, while users are running, spinning, or doing other activities, specialized motion detection equipment is used to record physiological information such as their heart rate. Second, by tracking the user's exercise records, rule-based exercise health recommendations are provided. However, this existing technology solution has the following drawbacks: First, it lacks a portable data recording function; and third, it fails to integrate big data technology to provide users with exercise risk assessment and analysis. Summary of the Invention

[0005] The present invention aims to overcome the shortcomings of existing methods by proposing a wearable device-based exercise risk assessment method. The present invention addresses the following main issues: 1. Existing solutions lack professionalism in exercise management, collect low amounts of data, are not conducive to portable record storage, and cannot achieve accurate risk assessment; 2. Existing solutions do not fully utilize big data analysis and fusion, resulting in low scalability and an inability to provide well-interpreted exercise recommendations.

[0006] In order to solve the above problems, the present invention proposes a sports risk assessment method based on a wearable device, which includes:

[0007] Authorize to obtain the user's initial data set, including the user's physiological indicators and environmental indicators, and store them on the server for model training to obtain a trained XGBoost model;

[0008] The user completes the collection of various physiological indicators based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client;

[0009] The client queries the environmental indicators in the corresponding initial data set online according to the current time, and splices the environmental indicators with the physiological indicator information according to the timestamp to obtain the total user data and transmits it to the server;

[0010] The server extracts and optimizes features of the total user data to obtain multi-source fusion features;

[0011] The server uses the trained XGBoost model to estimate the motion risk of the multi-source fusion features to obtain a risk estimation result, and calculates the importance of different features based on the number of features used in subtree splitting in the model;

[0012] The client obtains the risk estimation result from the server, performs a visual display based on the total user data, and generates suggestions for avoiding sports risks based on the importance of different features;

[0013] After the user receives the risk estimation result and the suggestion for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training for iterative updating.

[0014] Preferably, the authorization obtains the user's initial data set, including the user's physiological indicators and environmental indicators, and stores them on the server side, performs model training, and obtains a trained XGBoost model, specifically:

[0015] Obtain the physiological indicators, environmental indicators and corresponding sports injury information of users with different degrees of sports injuries through hospital authorization as the initial data set;

[0016] The physiological indicators of the initial dataset are collected within a given time period. The changing trends of the physiological indicators of a given user along with the timestamps, as well as the user's sports injury conditions, are used as model training labels.

[0017] The environmental indicators of the initial dataset are collected within a given time period to collect the indicator information of the sports environment of a given user, mainly including the user's location, air quality, air humidity, weather conditions, etc.

[0018] Based on the given data set, it is used as the user's initial data set and stored in the cloud server. The XGBoost model on the training server is updated to obtain the trained XGBoost model.

[0019] Preferably, the user completes the collection of various physiological indicator information based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client, specifically:

[0020] User physiological indicators collected include: heart rate, pulse, body temperature, exercise volume, exercise frequency, fat mass, blood oxygen saturation, skin conductance, red blood cell count, etc.

[0021] Establish a Bluetooth connection and send the collected physiological indicators at each timestamp to the client.

[0022] Preferably, the server extracts and optimizes features of the total user data to obtain multi-source fusion features, specifically:

[0023] The server extracts smooth features of k time periods using a time window weighted approach based on the obtained physiological indicator information of each wearable device as the timestamp changes. The resulting hierarchical features are:

[0024]

[0025] where f k is the extracted hierarchical feature, g k is the original feature, wk+i is the set time weight, T is the total time window size;

[0026] Based on the physiological indicators of each wearable device, the k smooth features extracted are combined with the historical data obtained from the initial data set for standardization, and the standard features are obtained as follows:

[0027]

[0028] in, is the mean of the kth feature in the historical data, σ k is the standard deviation of the kth feature in the historical data;

[0029] The physiological indicators thus obtained are binned and segmented according to a specific distribution, and the Pearson correlation coefficient is used to perform optimization in the data under the source of each wearable device to obtain the optimal features;

[0030] The optimal features obtained from each source are spliced ​​together to obtain multi-source fusion features, which are used as input features of the trained XGBoost model.

[0031] Preferably, after the user receives the risk estimation result and the suggestion for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training and iteratively updated, specifically:

[0032] The user client records the user's feedback on the sports risk estimation result and uploads the feedback to the server, which generates a label y i , and combined with the previous user input data x i , generate a new sample (x i ,y i ), add it to the model training data, iteratively update the original XGBoost model, the iterative formula is:

[0033]

[0034] The entire XGBoost model generates t trees in total, f t is the prediction result of the t-th tree, l() is the loss function, where g i and h i is the loss function l with respect to x i The coefficient of the second-order Taylor expansion, Ω() is the regularization constraint on the leaf nodes of the tree.

[0035] Accordingly, the present invention also provides a sports risk assessment system based on a wearable device, comprising:

[0036] The indicator information collection unit is used to obtain the user's initial data set for model training and complete the collection of various physiological indicator information and environmental indicator information;

[0037] The feature extraction and optimization unit is used to extract and optimize the features of the collected data to obtain multi-source fusion features as input features;

[0038] The data visualization unit is used to estimate sports risks, generate suggestions for avoiding sports risks, and perform visual display based on the total user data.

[0039] The implementation of the present invention has the following beneficial effects:

[0040] First, this invention proposes a wearable device-based exercise risk assessment method. This method uses smart wearable devices to collect physiological information related to human exercise, uploads it to a server for big data analysis and exercise risk assessment, and transmits it back to the smart device to provide risk assessment results and exercise recommendations. Second, different wearable device sensors monitor changes in physiological signals such as blood oxygen concentration, temperature, blood pressure, heart rate, and fat content. After heterogeneous data fusion in the cloud, the data is modeled using XGBoost and analysis techniques to assess exercise risk for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is an overall flow chart of a method for sports risk assessment based on a wearable device according to an embodiment of the present invention;

[0042] Figure 2 This is a structural diagram of a sports risk assessment system based on a wearable device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical inventions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Figure 1 This is a general flow chart of a method for sports risk assessment based on a wearable device according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0045] S1, authorizes the acquisition of the user's initial data set, including the user's physiological indicators and environmental indicators, and stores them on the server for model training to obtain a trained XGBoost model;

[0046] S2, the user completes the collection of various physiological indicators based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client;

[0047] S3, the client queries the environmental indicators in the corresponding initial data set online according to the current time, splices the environmental indicators with the physiological indicator information according to the timestamp, obtains the total user data and transmits it to the server;

[0048] S4, the server extracts and optimizes features from the total user data to obtain multi-source fusion features;

[0049] S5, the server uses the trained XGBoost model to estimate the motion risk of the multi-source fusion feature to obtain a risk estimation result, and calculates the importance of different features based on the number of features used in subtree splitting in the model;

[0050] S6, the client obtains the risk estimation result from the server, combines it with the total user data for visualization, and generates suggestions for avoiding sports risks based on the importance of the different features;

[0051] S7, after the user receives the risk estimation result and the suggestion for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training for iterative update.

[0052] Step S1 is as follows:

[0053] S1-1: Prepare the physiological indicator information of the dataset. Within a given time period, collect the changing trend of the physiological indicators of a given user with the timestamp, as well as the user's sports injury status as a model training label.

[0054] S1-2: Prepare the environmental metrics of the dataset and collect metrics about the environment over a given period of time. Based on the given dataset, update the XGBoost model on the training server.

[0055] Step S2 is as follows:

[0056] S2-1: The user uses multiple wearable devices to collect various physiological indicators, including heart rate, pulse, body temperature, exercise volume, exercise frequency, fat mass, blood oxygen saturation, skin conductance, and red blood cell count. A Bluetooth connection is established with the user's client, and the collected physiological indicators at each time stamp are sent to the client.

[0057] S2-2: The client queries the corresponding environmental indicator information online based on the current timestamp, which mainly includes the user's location, air quality, air humidity, weather conditions, etc.

[0058] Step S3 is as follows:

[0059] S3-1: The client concatenates the physiological indicator information and environmental indicator information obtained by S1 and S2 according to the timestamp as the total user data and transmits it to the server.

[0060] Step S4 is as follows:

[0061] S4-1: The server extracts smooth features of k time periods using a time window weighted approach based on the physiological indicator information of each wearable device obtained in S2 as the timestamp changes. Specifically, the resulting hierarchical features are:

[0062]

[0063] where f k is the extracted hierarchical feature, g k is the original feature, w k+i is the set time weight, and T is the total time window size.

[0064] S4-2: Based on the physiological indicators of each wearable device, the k features extracted by S4-1 are combined with the historical data obtained by S1 for standardization, and the new features are obtained as follows:

[0065]

[0066] in, is the mean of the kth feature in the historical data, σ k is the standard deviation of the kth feature in the historical data.

[0067] S4-3: The physiological indicators obtained in S4-2 are binned and segmented according to specific distributions to expand the data's expressive power. The resulting data is then optimized using the Pearson correlation coefficient within the data source for each wearable device. Specifically, highly correlated features are removed, preserving the differentiated expressive power of the features.

[0068] S4-4: Based on S4-3, the optimal features for each source are obtained. Since they mainly exist in the form of multiple columns of optimal features at this time, they are spliced ​​together and used as the input features of the model.

[0069] Step S5 is as follows:

[0070] S5-1: On the server side, use the XGBoost model trained in S1 to estimate motion risk using the multi-source fusion features obtained in S4. Calculate the importance of different features based on the number of features used in the subtree splitting in XGBoost.

[0071] Step S6 is as follows:

[0072] S6-1: The user client obtains the motion risk assessment results obtained in S5 from the server and visualizes them in combination with the data obtained in S2 and S3. The user client obtains the feature importance obtained in S5 from the server and generates motion risk avoidance recommendations based on this.

[0073] Step S7 is as follows:

[0074] S7-1: The user client records the user's feedback on the sports risk assessment results and uploads the feedback to the server. The server generates a label y based on the user feedback obtained. i , and combined with the previous user input data x i , generate a new sample (x i ,y i ).

[0075] S7-2: The new sample (x i ,y i ) is added to the training data of S1 to update the original XGBoost model. The model is iteratively updated according to the following formula:

[0076]

[0077] The entire XGBoost model generates t trees in total, f t is the prediction result of the t-th tree, l() is the loss function, where g i and h i is the loss function l with respect to x i The coefficient of the second-order Taylor expansion, Ω() is the regularization constraint on the leaf nodes of the tree.

[0078] Accordingly, the present invention also provides a sports risk assessment system based on wearable devices, such as Figure 2 As shown, including:

[0079] The indicator information collection unit 1 is used to obtain the user's initial data set for model training and complete the collection of various physiological indicator information and environmental indicator information.

[0080] Specifically, the indicator information collection unit obtains the physiological indicators, environmental indicators and corresponding sports injury information of users with different degrees of sports injuries through hospital authorization as the initial data set. The user completes the collection of various physiological indicator information based on wearable devices, including: heart rate, pulse, body temperature, exercise volume, exercise frequency, fat mass, blood oxygen saturation, skin conductance, red blood cell count, etc., and sends the collected physiological indicators at each timestamp to the client.

[0081] The feature extraction and optimization unit 2 is used to extract and optimize the features of the collected data to obtain multi-source fusion features as input features.

[0082] Specifically, the feature extraction optimization unit, the server extracts smooth features of k time periods using a time window weighted method based on the changes in the physiological indicator information of each wearable device as the timestamp changes, and obtains hierarchical features. Based on the physiological indicators of each wearable device, the k smooth features extracted are respectively combined with the historical data obtained from the initial data set for standardization to obtain standard features, and the various physiological indicators obtained in this way are binned and segmented according to a specific distribution. The Pearson correlation coefficient is used to perform optimization in the data under the source of each wearable device to obtain optimized features, and the optimized features obtained under each source are spliced ​​to obtain multi-source fusion features as the input features of the trained XGBoost model.

[0083] The data visualization unit 3 is used to estimate sports risks, generate suggestions for avoiding sports risks, and perform visual display in combination with the total user data.

[0084] Specifically, the data visualization unit and the client obtain the risk prediction results from the server, and perform a visual display in combination with the total user data, and generate suggestions for avoiding sports risks based on the importance of the different features. After the user receives the risk prediction results and the suggestions for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training for iterative updates.

[0085] Therefore, the present invention utilizes three components: a user's wearable device, a user client, and a cloud server to achieve wearable device data collection, integration, heterogeneous data fusion, and exercise risk assessment. First, the wearable device collects the user's physiological characteristics, primarily utilizing wearable device sensor modules such as heart rate, pulse, body temperature, exercise volume, exercise frequency, fat mass, blood oxygen saturation, skin conductance, and red blood cell count. Second, the client integrates the indicators collected by different wearable devices and interacts with the user at the terminal. The client collects environmental information such as climate, temperature, air quality, and location through network queries. The client then connects to the server, uploads the indicator and environmental information, and visualizes the exercise risk assessment results calculated by the server to the user. The cloud server then aggregates the long-term physiological characteristic curves collected by the user's various wearable devices and combines them with massive amounts of user data for data processing such as correlation analysis and binning and discrete coding. Ultimately, it filters, transforms, optimizes, and fuses the multi-source data to extract more effective data features. Finally, exercise risk assessment is performed based on the multiple comprehensive features generated by data fusion. This assessment model primarily uses the XGBoost algorithm. By integrating multiple decision trees to learn the correlations between massive amounts of user data, XGBoost achieves a highly interpretable sports risk assessment method. This interpretable analysis can then provide users with specific recommendations for managing sports risks.

[0086] The above is a detailed introduction to a sports risk assessment method and system based on wearable devices provided by an embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A sports risk assessment method based on wearable devices, characterized in that: The method comprises: Authorize the user's initial data set, including the user's physiological indicators and environmental indicators, and store it on the server for model training to obtain a trained XGBoost model; The user completes the collection of various physiological indicators based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client; The client queries the environmental indicators in the corresponding initial data set online according to the current time, and splices the environmental indicators with the physiological indicator information according to the timestamp to obtain the total user data and transmits it to the server; The server extracts and optimizes features of the total user data to obtain multi-source fusion features; The server uses the trained XGBoost model to estimate the motion risk of the multi-source fusion features to obtain a risk estimation result, and calculates the importance of different features based on the number of features used in subtree splitting in the model; The client obtains the risk estimation result from the server, performs a visual display based on the total user data, and generates suggestions for avoiding sports risks based on the importance of different features; After receiving the risk estimation results and the suggestions for avoiding sports risks, the user records his feedback and uploads it to the server as supplementary data for model training and iterative updates; Specifically, the authorization obtains the user's initial data set, including the user's physiological indicators and environmental indicators, and stores them on the server side, performs model training, and obtains a trained XGBoost model, specifically: Obtain the physiological indicators, environmental indicators and corresponding sports injury information of users with different degrees of sports injuries through hospital authorization as the initial data set; The physiological indicators of the initial dataset are collected within a given time period. The changing trends of the physiological indicators of a given user along with the timestamps, as well as the user's sports injury conditions, are used as model training labels. The environmental indicators of the initial dataset are indicators of a given user's sports environment collected within a given time period, including the user's location, air quality, and air humidity; Based on the given data set, it is used as the user's initial data set and stored in the cloud server. The XGBoost model on the training server is updated to obtain the trained XGBoost model. Specifically, the user completes the collection of various physiological indicator information based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client, specifically: User physiological indicators collected include: heart rate, pulse, body temperature, exercise volume, fat mass, blood oxygen saturation, skin conductance, and red blood cell count; Establish a Bluetooth connection and send the collected physiological indicators at each timestamp to the client; Specifically, the server extracts and optimizes the features of the total user data to obtain multi-source fusion features, specifically: The server extracts smooth features of k time periods using a time window weighted approach based on the obtained physiological indicator information of each wearable device as the timestamp changes. The resulting hierarchical features are: where f k is the extracted hierarchical feature, g k is the original feature, w k+i is the set time weight, T is the total time window size; Based on the physiological indicators of each wearable device, the k smooth features extracted are combined with the historical data obtained from the initial data set for standardization, and the standard features are obtained as follows: in, is the mean of the kth feature in the historical data, σ k is the standard deviation of the kth feature in the historical data; The physiological indicators thus obtained are binned and segmented according to a specific distribution, and the Pearson correlation coefficient is used to perform optimization in the data under the source of each wearable device to obtain the optimal features; The optimal features obtained from each source are spliced ​​to obtain multi-source fusion features as input features of the trained XGBoost model; Specifically, after the user receives the risk estimation result and the suggestion for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training and iteratively updated, specifically: The user client records the user's feedback on the sports risk estimation result and uploads the feedback to the server, which generates a label y i , and combined with the previous user input data x i , generate a new sample (x i ,y i ), add it to the model training data, iteratively update the original XGBoost model, the iterative formula is: The entire XGBoost model generates t trees in total, f t is the prediction result of the t-th tree, l() is the loss function, where g i and h i is the loss function l with respect to x i The coefficient of the second-order Taylor expansion, Ω() is the regularization constraint on the leaf nodes of the tree.

2. A sports risk assessment system based on wearable devices, characterized in that: The system comprises: The indicator information collection unit is used to authorize the acquisition of the user's initial data set, including the user's physiological indicators and environmental indicators, and store them on the server side for model training to obtain a trained XGBoost model; the user completes the collection of various physiological indicator information based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client; the client queries the environmental indicators in the corresponding initial data set based on the current time, splices them with the physiological indicator information based on the timestamp, obtains the total user data, and transmits it to the server side; A feature extraction and optimization unit is used for the server to extract and optimize features from the total user data to obtain multi-source fusion features; A data visualization unit is configured to enable the server to use the trained XGBoost model to estimate the motion risk of the multi-source fusion features, obtain a risk estimation result, and calculate the importance of different features based on the number of features used in subtree splitting in the model; the client obtains the risk estimation result from the server, performs a visual display in combination with the total user data, and generates motion risk avoidance suggestions based on the importance of the different features; after receiving the risk estimation result and the motion risk avoidance suggestions, the user's feedback is recorded and uploaded to the server as supplementary data for model training for iterative updating; Specifically, the indicator information collection unit is authorized to obtain the user's initial data set, including the user's physiological indicators and environmental indicators, and store them on the server side to perform model training to obtain a trained XGBoost model, specifically: Obtain the physiological indicators, environmental indicators and corresponding sports injury information of users with different degrees of sports injuries through hospital authorization as the initial data set; The physiological indicators of the initial dataset are collected within a given time period. The changing trends of the physiological indicators of a given user along with the timestamps, as well as the user's sports injury conditions, are used as model training labels. The environmental indicators of the initial dataset are indicators of a given user's sports environment collected within a given time period, including the user's location, air quality, and air humidity; Based on the given data set, it is used as the user's initial data set and stored in the cloud server. The XGBoost model on the training server is updated to obtain the trained XGBoost model. Specifically, in the indicator information collection unit, the user completes the collection of various physiological indicator information based on the wearable device, and the collected physiological indicators at each timestamp are sent to the client, specifically: User physiological indicators collected include: heart rate, pulse, body temperature, exercise volume, fat mass, blood oxygen saturation, skin conductance, and red blood cell count; Establish a Bluetooth connection and send the collected physiological indicators at each timestamp to the client; Specifically, in the feature extraction and optimization unit, the server extracts and optimizes the features of the total user data to obtain multi-source fusion features, specifically: The server extracts smooth features of k time periods using a time window weighted approach based on the obtained physiological indicator information of each wearable device as the timestamp changes. The resulting hierarchical features are: where f k is the extracted hierarchical feature, g k is the original feature, w k+i is the set time weight, T is the total time window size; Based on the physiological indicators of each wearable device, the k smooth features extracted are combined with the historical data obtained from the initial data set for standardization, and the standard features are obtained as follows: in, is the mean of the kth feature in the historical data, σ k is the standard deviation of the kth feature in the historical data; The physiological indicators thus obtained are binned and segmented according to a specific distribution, and the Pearson correlation coefficient is used to perform optimization in the data under the source of each wearable device to obtain the optimal features; The optimal features obtained from each source are spliced ​​to obtain multi-source fusion features as input features of the trained XGBoost model; Specifically, in the data visualization unit, after the user receives the risk estimation results and the suggestions for avoiding sports risks, the user's feedback is recorded and uploaded to the server as supplementary data for the model training and iteratively updated, specifically: The user client records the user's feedback on the sports risk estimation result and uploads the feedback to the server, which generates a label y i , and combined with the previous user input data x i , generate a new sample (x i ,y i ), add it to the model training data, iteratively update the original XGBoost model, the iterative formula is: The entire XGBoost model generates t trees in total, f t is the prediction result of the t-th tree, l() is the loss function, where g i and h i is the loss function l with respect to x i The coefficient of the second-order Taylor expansion, Ω() is the regularization constraint on the leaf nodes of the tree.

Images