A heart rate variability detection system based on scatter plot
By designing a heart rate variability detection system based on scatter plots, constructing and analyzing heart rate variability scatter plots, and combining them with a body state classification model, the system solves the problem of insufficient accuracy in existing electrocardiogram signal analysis and achieves efficient classification of body states.
Patent Information
- Application Number
- CN202410630397.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-21
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-05-21
AI Technical Summary
Existing technologies are insufficient for efficiently and accurately classifying the physical condition of a target object through electrocardiogram signal analysis.
Design a heart rate variability detection system based on scatter plots, including a heart rate variability detection module, a scatter plot construction module, an analysis and acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module. By constructing a heart rate variability scatter plot and analyzing its features, combined with a pre-trained body state classification model, the system can classify body states.
It improves the accuracy and efficiency of electrocardiogram signal analysis, enables accurate identification of physical conditions, and enhances the objective analysis capabilities of data.
Smart Images

Figure CN118512160B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a heart rate variability detection system based on scatter plots, belonging to the field of scatter plot analysis technology. Background Technology
[0002] Heart rate variability (HRV) refers to the fluctuation of the RR interval in each heartbeat. Its mechanism involves the coordinated action of the sympathetic and vagus nerves, which control the pacing of the sinoatrial node. Specifically, it manifests as irregular variations in the length of each cardiac cycle. HRV signals contain a wealth of information about cardiovascular control and humoral regulation. Extracting and analyzing this information allows for a quantitative assessment of the balance between the cardiac sympathetic and vagus nerves. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a heart rate variability detection system based on scatter plots. By designing and constructing various modules, the system analyzes the patterns of electrocardiogram signals and efficiently and accurately classifies the physical state of the target object.
[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a heart rate variability detection system based on scatter plots, including a heart rate variability detection module, a scatter plot construction module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module;
[0005] Among them, the heart rate variability detection module is used to collect the electrocardiogram (ECG) signal of the target object, and based on the collected ECG signal, obtain the time interval data between adjacent heartbeat R waves within the corresponding preset detection time period;
[0006] The scatter plot construction module is used to construct data from the heart rate variability detection module. Based on a target coordinate system where both the positive x-axis and positive y-axis axes are durations, and the scales of the positive x-axis and positive y-axis axes are the same, the module uses the duration interval R between the nth heartbeat R-wave and the (n+1)th heartbeat R-wave as an example. n R n+1 The x-axis represents the time interval R between the (n+1)th heartbeat R-wave and the (n+2)th heartbeat R-wave. n+1 R n+2 Using the vertical axis as the ordinate, construct a scatter plot of heart rate variability of the target object in the target coordinate system, and then construct a scatter plot of heart rate variability of the target object corresponding to the preset detection time period;
[0007] The data acquisition module analyzes the heart rate variability scatter plot constructed by the scatter plot construction module. Starting from the origin of the target coordinate system, the reference ray is taken along the 45-degree direction between the positive horizontal axis and the positive vertical axis. The distance between the two outermost scatter points in the direction perpendicular to the reference ray is obtained sequentially. Then, the distance corresponding to each position in the direction of the reference ray is obtained.
[0008] The scatter feature acquisition module is used to analyze the spacing corresponding to each position along the reference ray direction in the heart rate variability scatter plot obtained by the data acquisition module, and to obtain the variation pattern of each spacing along the reference ray direction in the heart rate variability scatter plot.
[0009] The state analysis and inference module is used to infer the body state classification of the target object for a preset detection time period by taking the variation pattern of each interval along the reference ray in the heart rate variability scatter plot obtained by the scatter feature acquisition module, the age of the target object, and a pre-trained body state classification model with the variation pattern of each interval along the reference ray in the heart rate variability scatter plot and the age of the target object as input and the body state classification of the target object as output.
[0010] As a preferred technical solution of the present invention, it further includes a dynamic scatter plot construction module and an excitation state analysis module, wherein, based on the heart rate variability detection module, the time interval data between adjacent heartbeat R waves in each consecutive preset detection time segment corresponding to the target object are collected, and the heart rate variability scatter plot corresponding to each consecutive preset detection time segment is constructed by the scatter plot construction module, and then the dynamic scatter plot construction module constructs a dynamic change map of heart rate variability scatter plot corresponding to each consecutive preset detection time segment based on the heart rate variability scatter plot corresponding to each consecutive preset detection time segment;
[0011] The excitation state analysis module is used to analyze the dynamic change graph of heart rate variability scatter points constructed by the dynamic scatter point construction module. Based on the principle that the closer the scatter point region is to the origin of the target coordinate system, the greater the corresponding excitation level, and the farther the scatter point region is from the origin of the target coordinate system, the smaller the corresponding excitation level, the module analyzes the position change of the scatter point region in the dynamic change graph of heart rate variability scatter points in the target coordinate system, and obtains the change of the excitation level of the target object with each consecutive preset detection duration.
[0012] As a preferred technical solution of the present invention, it further includes an abnormal scatter point detection module. The abnormal scatter point detection module is used to detect each scatter point in the heart rate variability scatter plot constructed by the scatter point construction module that has a distance greater than a preset distance upper limit between it and its adjacent scatter points, delete each scatter point, update the heart rate variability scatter plot, and then the analysis data acquisition module processes the heart rate variability scatter plot to obtain the distance corresponding to each position in the reference ray direction.
[0013] As a preferred embodiment of the present invention, the variation patterns of the intervals along the reference ray direction in the heart rate variability scatter plot obtained by the scatter feature acquisition module include the following four types:
[0014] (1) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The number of spacings between the first 0 spacing and the maximum value spacing is greater than the number of spacings between the second maximum value spacing and the 0 spacing. The proportion of this difference to the total number of spacings is greater than the upper limit of the preset difference proportion.
[0015] (2) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The difference between the number of spacings between the first 0 spacing and the maximum spacing and the number of spacings between the maximum spacing and the 0 spacing in the second segment is less than the lower limit of the preset difference ratio, and the upper limit of the preset difference ratio is greater than the lower limit of the preset difference ratio.
[0016] (3) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of the scattered points is located to the lower left of the critical line.
[0017] (4) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of scattered points is distributed from the lower left of the critical line to the upper right of the critical line.
[0018] As a preferred technical solution of the present invention, the pre-trained body state classification model is obtained by a sample data acquisition module, a sample construction module, and a model training module.
[0019] The sample data acquisition module is used to process the sample objects corresponding to the preset number of body state categories in sequence through the heart rate variability detection module, scatter plot construction module, analysis data acquisition module, and scatter plot feature acquisition module to obtain the variation pattern of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to the preset detection time period of the sample object.
[0020] The sample construction module is used to combine the variation patterns of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to each sample object obtained by the sample data acquisition module, and combine the age of each sample object and the corresponding body state classification of each sample object to construct the sample data corresponding to each sample object.
[0021] The model training module is used to train a body state classification model based on the sample data corresponding to each sample object constructed by the sample construction module. The inputs are the sample object's age and the variation pattern of each interval along the reference ray in the scatter plot of the sample object's corresponding heart rate variability. The output is the body state classification of the sample object.
[0022] As a preferred technical solution of the present invention: the preset classification network includes random forest network, support vector machine network, extreme gradient boosting network, neural network, and decision tree network.
[0023] The heart rate variability detection system based on scatter plots described in this invention has the following technical advantages compared with existing technologies:
[0024] (1) The heart rate variability detection system based on scatter plot designed in this invention includes a heart rate variability detection module, a scatter plot construction module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module. For the time interval data between adjacent heartbeat R waves corresponding to the electrocardiogram signal, the system analyzes the data using a scatter plot under the designed coordinates. Based on the spacing corresponding to each position along the reference ray direction in the heart rate variability scatter plot, the system obtains the variation law of each spacing along the reference ray direction in the heart rate variability scatter plot. Finally, combined with the age of the target object, the system applies a pre-trained body state classification model to infer the body state classification of the target object corresponding to the preset detection time period. The design scheme is based on the features of the heart rate variability scatter plot and combined with big data analysis, which can accurately identify the body state and improve the accuracy and efficiency of objective data analysis. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the architecture of the heart rate variability detection system based on scatter plots designed in this invention;
[0026] Figure 2 This is an embodiment diagram of the heart rate variability detection system based on scatter plots designed in this invention, showing the variation pattern at different intervals;
[0027] Figure 3 This invention relates to a heart rate variability detection system based on scatter plots, which uses dynamic changes in heart rate variability scatter plots. Detailed Implementation
[0028] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0029] Extensive data analysis revealed that the RR interval scatter plot is nonlinear (chaotic) and can intuitively express the nonlinear dynamic characteristics of HRV, serving both qualitative and quantitative purposes. The scatter plot aggregates heart rhythms of the same nature into an "attractor." When the origin or location of the heart rhythm changes, new "attractors" are generated, thus altering the shape of the scatter plot. This provides corresponding evidence for cardiac diagnosis. Heart rate variability time series have chaotic characteristics, and the scatter plot represents the RR interval time series in a two-dimensional plane in a scatter plot format. The distribution pattern and density of the scatter points reflect the nonlinear motion pattern of the heart.
[0030] Furthermore, extensive data analysis revealed that the scatter plots corresponding to normal individuals... Figure 1 Scatter plots typically resemble comets, meaning the plot is narrower closer to the origin and wider further away. Furthermore, scatter plots located distal to the 45° line indicate increased vagal activity (slower heart rate), while scatter plots proximal to the 45° line indicate increased sympathetic activity (faster heart rate). These characteristics can provide a basis for diagnosing arrhythmias and assessing autonomic nervous system function.
[0031] Based on the findings from the analysis of a large amount of data, the present invention designs a heart rate variability detection system based on scatter plots, such as... Figure 1 As shown, it includes a heart rate variability detection module, a scatter plot construction module, an abnormal scatter plot detection module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module.
[0032] The heart rate variability detection module is used to collect the electrocardiogram (ECG) signal of the target object and, based on the collected ECG signal, obtain the time interval data between adjacent heartbeat R waves within the corresponding preset detection time period.
[0033] The scatter plot construction module is used to construct data from the heart rate variability detection module. Based on a target coordinate system where both the positive x-axis and positive y-axis axes are durations, and the scales of the positive x-axis and positive y-axis axes are the same, the module uses the duration interval R between the nth heartbeat R-wave and the (n+1)th heartbeat R-wave as an example. n R n+1 The x-axis represents the time interval R between the (n+1)th heartbeat R-wave and the (n+2)th heartbeat R-wave. n+1 R n+2 Using the vertical axis, a scatter plot is constructed in the target coordinate system, and then a scatter plot of heart rate variability corresponding to the target object for the preset detection time period is constructed.
[0034] The abnormal scatter plot detection module is used to detect scatter plots of heart rate variability constructed by the scatter plot construction module. It detects scatter plots whose distance from adjacent scatter plots is greater than the preset distance limit, deletes each scatter plot, and updates the heart rate variability scatter plot.
[0035] The data acquisition analysis module is used to process the updated heart rate variability scatter plot after the abnormal scatter detection module. Taking the origin of the target coordinate system as the starting point and the 45-degree direction between the positive horizontal axis and the positive vertical axis as the reference ray, the distance between the two outermost scatter points in the direction perpendicular to the reference ray is obtained sequentially. Then, the distance corresponding to each position in the direction of the reference ray is obtained.
[0036] The scatter feature acquisition module is used to analyze the spacing corresponding to each position along the reference ray direction in the heart rate variability scatter plot obtained by the data acquisition module, and to obtain the variation law of each spacing along the reference ray direction in the heart rate variability scatter plot. In practical applications, through the analysis of a large amount of data, it was found that the variation law here mainly presents the following four types.
[0037] (1) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The number of spacings between the first 0 spacing and the maximum value spacing is greater than the number of spacings between the second maximum value spacing and the 0 spacing. The proportion of this difference to the total number of spacings is greater than the upper limit of the preset difference proportion.
[0038] (2) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The difference between the number of spacings between the first 0 spacing and the maximum spacing and the number of spacings between the maximum spacing and the 0 spacing in the second segment is less than the lower limit of the preset difference ratio, and the upper limit of the preset difference ratio is greater than the lower limit of the preset difference ratio.
[0039] (3) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of the scattered points is located to the lower left of the critical line.
[0040] (4) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of scattered points is distributed from the lower left of the critical line to the upper right of the critical line.
[0041] like Figure 2 As shown, the scatter distribution in Figure 1 represents normal sinus rhythm; the scatter distribution in Figure 2 represents sinus bradycardia in coronary artery disease; the scatter distribution in Figure 3 represents the inherent sinus rhythm in diabetes; the scatter distribution in Figures 4 and 5 represents sinus rhythm in acute myocardial infarction; and the scatter distribution in Figure 6 represents sinus tachycardia-bradycardia syndrome.
[0042] In practical applications, when corresponding to the first and second variation patterns, that is, the spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing, the maximum spacing value SD1 can reflect the "rapid change" of vagal tone; the length SD2 of the segment on the reference ray corresponding to the change from 0 spacing to 0 spacing reflects the "slow change" of sympathetic tone; the ratio of SD1 to SD2, <0.5 indicates that the significance of predicting ischemia is uncertain; >0.5 indicates that the prediction of ischemia is significant; >1 indicates that the incidence of ischemia is 100%.
[0043] The state analysis and inference module is used to infer the body state classification of the target object for a preset detection time period by taking the variation pattern of each interval along the reference ray in the heart rate variability scatter plot obtained by the scatter feature acquisition module, the age of the target object, and a pre-trained body state classification model with the variation pattern of each interval along the reference ray in the heart rate variability scatter plot and the age of the target object as input and the body state classification of the target object as output.
[0044] In practical applications, the pre-trained body state classification model is obtained from the sample data acquisition module, the sample construction module, and the model training module.
[0045] The sample data acquisition module is used to process a preset number of sample objects corresponding to each preset body state category through the heart rate variability detection module, scatter plot construction module, analysis data acquisition module, and scatter plot feature acquisition module in sequence to obtain the variation pattern of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to the preset detection time period of the sample object.
[0046] The sample construction module is used to combine the variation patterns of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to each sample object obtained by the sample data acquisition module, and the age and corresponding physical state classification of each sample object to construct the sample data corresponding to each sample object.
[0047] The model training module is used to train a body state classification model based on the sample data corresponding to each sample object constructed by the sample construction module. The model takes the sample object's age and the variation pattern of each interval along the reference ray in the heart rate variability scatter plot of the sample object as input and the body state classification of the sample object as output. The model training module is used to train a preset classification network such as random forest network, support vector machine network, extreme gradient boosting network, neural network network, or decision tree network.
[0048] Based on the above design, which includes a heart rate variability detection module, a scatter plot construction module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module, in practical applications, a dynamic scatter plot construction module and an excitation state analysis module can be further designed and introduced. Specifically, the heart rate variability detection module collects the duration interval data between adjacent heartbeat R waves within each consecutive preset detection duration segment for the target object, and the scatter plot construction module constructs heart rate variability scatter plots corresponding to each consecutive preset detection duration segment. Then, the dynamic scatter plot construction module constructs a dynamic change map of heart rate variability scatter plots corresponding to each consecutive preset detection duration segment, based on the heart rate variability scatter plots.
[0049] The excitation state analysis module is used to analyze the dynamic change graph of heart rate variability scatter points constructed by the dynamic scatter point construction module. Based on the principle that the closer the scatter point region is to the origin of the target coordinate system, the greater the corresponding excitation level, and the farther the scatter point region is from the origin of the target coordinate system, the smaller the corresponding excitation level, the module analyzes the position change of the scatter point region in the dynamic change graph of heart rate variability scatter points in the target coordinate system, and obtains the change of the excitation level of the target object with each consecutive preset detection duration.
[0050] In practical applications, such as Figure 3 As shown in the figure, the scatter plot distribution areas corresponding to the nighttime period are different from those corresponding to the daytime period, and the level of excitement corresponding to the daytime period is obviously greater than that corresponding to the nighttime period.
[0051] The aforementioned technical solution designs a heart rate variability detection system based on scatter plots, including a heart rate variability detection module, a scatter plot construction module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module. It analyzes the time interval data between adjacent heartbeat R waves corresponding to electrocardiogram signals using a scatter plot in a designed coordinate system. Based on the spacing corresponding to each position along the reference ray direction in the heart rate variability scatter plot, it obtains the variation law of each spacing along the reference ray direction in the heart rate variability scatter plot. Finally, combined with the age of the target object, a pre-trained body state classification model is applied to infer the body state classification of the target object for a preset detection time period. This design scheme, based on the features of heart rate variability scatter plots and combined with big data analysis, can accurately identify body states and improve the accuracy and efficiency of objective data analysis.
[0052] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A heart rate variability detection system based on scatter plots, characterized in that: It includes a heart rate variability detection module, a scatter plot construction module, an analysis data acquisition module, a scatter plot feature acquisition module, and a state analysis and inference module; Among them, the heart rate variability detection module is used to collect the electrocardiogram (ECG) signal of the target object, and based on the collected ECG signal, obtain the time interval data between adjacent heartbeat R waves within the corresponding preset detection time period; The scatter plot construction module is used to construct data from the heart rate variability detection module. Based on a target coordinate system where both the positive x-axis and positive y-axis axes are durations, and the scales of the positive x-axis and positive y-axis axes are the same, the module uses the duration interval R between the nth heartbeat R-wave and the (n+1)th heartbeat R-wave as an example. n R n+1 The x-axis represents the time interval R between the (n+1)th heartbeat R-wave and the (n+2)th heartbeat R-wave. n+1 R n+2 Using the vertical axis as the ordinate, construct a scatter plot of heart rate variability of the target object in the target coordinate system, and then construct a scatter plot of heart rate variability of the target object corresponding to the preset detection time period; The data acquisition module analyzes the heart rate variability scatter plot constructed by the scatter plot construction module. Starting from the origin of the target coordinate system, the reference ray is taken along the 45-degree direction between the positive horizontal axis and the positive vertical axis. The distance between the two outermost scatter points in the direction perpendicular to the reference ray is obtained sequentially. Then, the distance corresponding to each position in the direction of the reference ray is obtained. The scatter feature acquisition module is used to analyze the spacing corresponding to each position along the reference ray direction in the heart rate variability scatter plot obtained by the data acquisition module, and to obtain the variation pattern of each spacing along the reference ray direction in the heart rate variability scatter plot. The state analysis and inference module is used to infer the body state classification of the target object for the target object based on the variation pattern of each interval along the reference ray in the heart rate variability scatter plot obtained by the scatter feature acquisition module, as well as the age of the target object. It uses a pre-trained body state classification model with the variation pattern of each interval along the reference ray in the heart rate variability scatter plot and the age of the target object as input and the body state classification of the target object as output. It also includes a dynamic scatter plot construction module and an excitation state analysis module. The heart rate variability detection module collects the time interval data between adjacent heartbeat R waves in each consecutive preset detection time segment corresponding to the target object. The scatter plot construction module constructs the heart rate variability scatter plot corresponding to each consecutive preset detection time segment. Then, the dynamic scatter plot construction module constructs a dynamic change map of heart rate variability scatter plot corresponding to each consecutive preset detection time segment based on the heart rate variability scatter plot corresponding to each consecutive preset detection time segment. The excitation state analysis module is used to analyze the dynamic change graph of heart rate variability scatter points constructed by the dynamic scatter point construction module. Based on the principle that the closer the scatter point region is to the origin of the target coordinate system, the greater the corresponding excitation level, and the farther the scatter point region is from the origin of the target coordinate system, the lower the corresponding excitation level, the module analyzes the position change of the scatter point region in the dynamic change graph of heart rate variability scatter points in the target coordinate system, and obtains the change of the excitation level of the target object with each consecutive preset detection duration. It also includes an abnormal scatter point detection module, which is used to detect each scatter point in the heart rate variability scatter plot constructed by the scatter point construction module that has a distance greater than the preset distance upper limit between it and its adjacent scatter points, delete each scatter point, update the heart rate variability scatter plot, and then the analysis data acquisition module processes the heart rate variability scatter plot to obtain the distance corresponding to each position in the reference ray direction.
2. The heart rate variability detection system based on scatter plots according to claim 1, characterized in that: The variation patterns of each interval along the reference ray direction in the heart rate variability scatter plot obtained by the scatter feature acquisition module include the following four types: (1) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The number of spacings between the first 0 spacing and the maximum value spacing is greater than the number of spacings between the second maximum value spacing and the 0 spacing. The proportion of this difference to the total number of spacings is greater than the upper limit of the preset difference proportion. (2) The spacing changes from 0 spacing to gradually increase and then gradually decrease back to 0 spacing. The difference between the number of spacings between the first 0 spacing and the maximum spacing and the number of spacings between the second maximum spacing and the 0 spacing is less than the lower limit of the preset difference ratio, and the upper limit of the preset difference ratio is greater than the lower limit of the preset difference ratio. (3) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of the scattered points is located to the lower left of the critical line. (4) Define a critical line that passes through a preset position on the reference ray and is perpendicular to the reference ray. The spacing changes from 0 to gradually increase, and the range of scattered points is distributed from the lower left of the critical line to the upper right of the critical line.
3. The heart rate variability detection system based on scatter plots according to claim 1, characterized in that, The pre-trained body state classification model is obtained from the sample data acquisition module, the sample construction module, and the model training module. The sample data acquisition module is used to process the sample objects corresponding to the preset number of body state categories in sequence through the heart rate variability detection module, scatter plot construction module, analysis data acquisition module, and scatter plot feature acquisition module to obtain the variation pattern of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to the preset detection time period of the sample object. The sample construction module is used to combine the variation patterns of each interval along the reference ray direction in the heart rate variability scatter plot corresponding to each sample object obtained by the sample data acquisition module, and combine the age of each sample object and the corresponding body state classification of each sample object to construct the sample data corresponding to each sample object. The model training module is used to train a body state classification model based on the sample data corresponding to each sample object constructed by the sample construction module. The inputs are the sample object's age and the variation pattern of each interval along the reference ray in the scatter plot of the sample object's corresponding heart rate variability. The output is the body state classification of the sample object.
4. The heart rate variability detection system based on scatter plots according to claim 3, characterized in that: The preset classification network includes random forest network, support vector machine network, extreme gradient boosting network, neural network, and decision tree network.
