Respiratory system disease prediction system based on machine learning

Abstract

The invention discloses a respiratory system disease prediction system based on machine learning. The system comprises a data preparation module, a data preprocessing module, a feature extraction module, a respiratory system disease prediction model construction module, an analysis model output module and a report generation module. The invention belongs to the technical field of disease early warning, and particularly relates to a respiratory system disease prediction system based on machine learning, which adopts discretization processing on data, so that subsequent data analysis and model construction are more efficient and accurate, and the prediction accuracy is improved. Important basis and support are provided for deeply researching the influence mechanism of air pollutants on respiratory system diseases; multi-scale information particles are fused by adopting a multi-scale information fusion technology, significance of each scale is calculated by utilizing data distribution difference, and data features are better grasped; the ABC algorithm is adopted to optimize and trim the parameters of the tree, and the tree structure is accurately regulated and controlled, so that effective trimming and node setting of the tree are realized, and the minimum prediction error is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of disease early warning, and specifically is a respiratory disease prediction system based on machine learning. Background Art

[0002] The respiratory disease prediction system is a system that uses air quality data to explore the possible relationship between environmental factors and respiratory diseases, and provides early warning of diseases, public health decisions, and appropriate protective measures for specific areas. It aims to improve the prevention and control capabilities of respiratory diseases, better protect public health, identify potential risks in advance, and reduce the incidence and severity of respiratory diseases. However, the existing respiratory disease prediction system has technical problems such as difficulty in finding potential correlation patterns between air pollutants and patients with respiratory diseases, complex raw data, and difficulty in direct analysis; incomplete spatial information of air pollutant emission data, and difficulty in extracting valuable air pollutant emission data; and large errors in respiratory disease predictions, which mislead public health departments in judging the epidemic trend of the disease, thereby affecting the formulation of effective prevention and control strategies and measures. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a respiratory disease prediction system based on machine learning. In view of the technical problems that it is difficult to find the potential correlation pattern between air pollutants and patients with respiratory diseases, the original data is complex and difficult to directly analyze, the data is discretized to reduce the complexity of the data, so that the subsequent data analysis and model construction are more efficient and accurate, and provide important basis and support for in-depth research on the impact mechanism of air pollutants on respiratory diseases and the formulation of relevant environmental protection and health strategies; in view of the technical problems that the spatial information of air pollutant emission data is incomplete and it is difficult to extract valuable air pollutant emission data, the multi-scale information fusion technology is used to fuse the multi-scale information particles, and the significance of each scale is calculated by using the data distribution difference to better grasp the data characteristics; in view of the technical problems that there are large errors in the prediction of respiratory diseases, which mislead the public health department's judgment on the epidemic trend of the disease, thereby affecting the formulation of effective prevention and control strategies and measures, the ABC algorithm is used to optimize the tree parameters and prune them, construct a CART tree, and accurately regulate the tree structure to achieve effective pruning and node setting of the tree, so as to ensure the minimum prediction error.

[0004] The present invention provides a respiratory disease prediction system based on machine learning, comprising a data preparation module, a data preprocessing module, a feature extraction module, a respiratory disease prediction model building module, an analysis model output module and a report generation module;

[0005] The data preparation module specifically collects data on air pollutant emissions and the number of patients with respiratory diseases;

[0006] The data preprocessing module specifically adopts an unsupervised discretization technology based on an artificial neural network self-organizing map, trains the network through a competitive learning mechanism to distribute the air pollutant emission data to learn potential patterns and features, and maps and discretizes the air pollutant emission data according to clustering areas;

[0007] The feature extraction module specifically uses a multi-label and multi-scale learning strategy to mine potential multiple classification information, uses correlation differences to fit true labels, uses distribution differences to correct ratio deviations, builds a model for missing label multi-label data, and calculates significance;

[0008] The respiratory disease prediction model building module specifically builds a CART tree, uses the ABC algorithm to prune leaves and set split points, calculates the fitness unit of the solution vector, uses the training error as the fitness function, calculates the fitness of the solution vector, and ensures that the learning model prediction error is minimized;

[0009] The analysis model output module is specifically used to visualize data and display the model's prediction results of respiratory diseases in the form of intuitive charts;

[0010] The report generation module is specifically for presenting and explaining the prediction results in detail so that non-professionals can also effectively understand it and summarize the model as a whole.

[0011] Furthermore, in the data preparation module, the data preparation module specifically collects long-term series of air pollutant emission data and data on the number of patients with respiratory diseases during the same period, and the air pollutants include PM2.5, PM10, sulfur dioxide and nitrogen oxides.

[0012] Furthermore, in the data preprocessing module, a data cleaning unit and a data analysis processing unit are provided, and the data preprocessing module includes the following contents:

[0013] Data cleaning unit, which collects data on air pollutant emissions and the number of patients with respiratory diseases, removes noise, corrects erroneous data, and handles missing values;

[0014] The data analysis and processing unit uses unsupervised discretization technology based on artificial neural network self-organizing map to analyze and process air pollutant emission data, including the following:

[0015] Network initialization: determine the number of nodes and topological structure of the self-organizing graph network, and randomly initialize the node vector;

[0016] Training the network, inputting air pollutant emission data into the network, and using a competitive learning mechanism to gradually adapt the nodes to the distribution of air pollutant emission data, so as to learn the potential patterns and features in the air pollutant emission data;

[0017] Forming clusters, after training, the network nodes will form different cluster areas to reflect the similarity and correlation between air pollutant emission data;

[0018] Discrete mapping maps and discretizes the air pollutant emission data according to the clustering area to which it belongs in the network, and determines the discrete category corresponding to each air pollutant emission data. The discrete categories include discretization according to the concentration range of the pollutant and discretization according to the type of pollutant.

[0019] Furthermore, in the feature extraction module, there are provided a space conversion unit, a multi-scale information particle fusion unit, a true label fitting unit, a significance calculation unit of each scale and a feature classification unit. The feature extraction module includes the following contents:

[0020] The spatial transformation unit transforms a single data space into a multi-scale fuzzy granular space through a fuzzy neighborhood granularity recognition mechanism, and uses a multi-label multi-scale learning strategy to mine potential multi-classification information;

[0021] The multi-scale information particle units are integrated to construct a multi-scale fusion fuzzy granularity space to ensure the diversity and integrity of pollutant information in the air pollutant emission data space, and the value of the multi-label multi-scale fuzzy dependency function at the k scale is calculated. The formula used is as follows:

[0022] ;

[0023] In the formula, represents the value of the multi-label multi-scale fuzzy dependency function at scale k, B represents the air pollutant emission category dataset, U represents the full set of data, Represents the i-th element in the data set, i represents the index of the element in the data set, It represents the correlation degree between the element and the air pollutant emission category dataset B at the k scale, Represents a fuzzy label set and reconstructs a new label set based on the data with missing labels;

[0024] Fitting the true label unit follows the principle that samples with greater correlation in the multi-scale fusion granularity space have more similar labels, and uses the difference in the correlation ratio between labeled and unlabeled samples in the sample granularity structure to fit the true label;

[0025] Calculate the significance units of each scale, use the distribution difference in the data to correct the ratio deviation, and build a multi-scale fuzzy rough set model for multi-label data with missing labels to calculate the significance of each scale. The formula used is as follows:

[0026] ;

[0027] In the formula, represents the significance at scale k, represents the data distribution difference at this scale, and f represents the functional relationship for calculating the significance based on the distribution difference;

[0028] The feature classification unit defines the multi-scale fusion fuzzy uncertainty metric, explores the correlation between features and fuzzy label sets, the redundancy and interactivity between features, designs corresponding feature evaluation criteria, selects discriminant features for classification, and thus extracts valuable air pollutant emission data for further analysis.

[0029] Furthermore, in the module for constructing a respiratory disease prediction model, an ABC algorithm optimization tree unit, a CART construction unit, an optimization variable unit, and a fitness unit for calculating a solution vector are provided. The module for constructing a respiratory disease prediction model includes the following contents:

[0030] The ABC algorithm optimizes the tree unit and uses the ABC algorithm to optimize and prune the parameters of the tree. The tree components include decision nodes and leaf nodes. Each decision node represents one of the constraints that constitute the tree rules. Based on the constraint, the input sample is analyzed according to the condition of one of the decision variables, and the fitness function of the ETC-related indicators is calculated. The formula used is as follows:

[0031] ;

[0032] In the formula, represents the fitness function of ETC-related indicators, T represents the specified target vector, C represents the function for calculating the correlation coefficient, N1 represents the number of samples, , and Represents a specific sample parameter, i, j and a are the indexes of the sample;

[0033] CART construction unit,The CART construction process consists of selecting input variables and,determining the split points of these variables until a suitable tree is created,which is done using a greedy algorithm to minimize the cost function.,In each splitting step, the point with the lowest cost is selected;

[0034] The CART model is used for classification, and the Gini impurity index is used to determine the cost function for building the tree. The formula used is as follows:

[0035] ;

[0036] In the formula, Gini represents Gini impurity, which is used to measure the degree of confusion of the data set, C represents the number of target classes, that is, the total number of different categories to be classified, and p c1 Indicates the proportion of training instances marked as category c1 in the current node;

[0037] Optimize variable units. After building the initial CART model, use the ABC algorithm to prune leaves and set split points. For a decision tree with N decision nodes and M leaf nodes, optimize variables, including the following:

[0038] N real variables, each variable corresponds to a node of the decision tree, and the value of the variable represents the split point of the node. Since all input variables of the problem are normalized using the formula, the search boundaries of all variables are (0, 1);

[0039] M binary variables, each binary variable corresponds to a leaf node of the tree. In the binary variables, a value of zero indicates pruning the branch corresponding to the leaf node in the tree structure, and a value of one indicates the maintenance of the node in the tree structure;

[0040] Each solution vector in the ABC algorithm has a heterogeneous structure. In each solution vector, the first N elements are used to determine the split point of the decision node, and the last M elements specify the deletion and retention of leaf nodes in the tree structure;

[0041] The fitness unit of the solution vector is calculated. Each solution vector is applied to the initial tree structure. The fitness of the solution vector is calculated based on the resulting tree. The training error criterion is used as the fitness function in the ABC algorithm. The validation sample is applied to the tree modified by the solution vector, and the output label of the tree is compared with the actual label. The fitness of the solution vector is calculated using the formula. The formula used is as follows:

[0042] ;

[0043] Where Fit represents the fitness of the solution vector, E represents the number of validation instances where the output of the tree is different from the actual label of the instance, and N represents the total number of real variables. By using the proposed structure for the solution vector, the decision nodes in the learning model are effectively set while pruning the tree. The training error criterion is used as the fitness function in the optimization algorithm to ensure that the prediction error in the learning model is minimized.

[0044] Furthermore, in the analysis model output module, data visualization is performed to display the model's prediction results for respiratory diseases in the form of intuitive charts, including a bar chart of the prediction ratios of different disease categories and a trend chart of individual patient prediction results, to analyze the different changing trends and patterns of the prediction results under different air pollutant emission data.

[0045] Furthermore, in the report generation module, the prediction results are presented specifically, including the numerical values ​​of various indicators, chart displays and case analysis, and the prediction results are displayed and explained so that non-professionals can better understand them, summarizing the overall performance and effectiveness of the model and its value in predicting respiratory diseases.

[0046] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0047] (1) In order to solve the technical problems of finding potential correlation patterns between air pollutants and patients with respiratory diseases and the complexity of original data, which are difficult to directly analyze, we adopt the method of discretizing the data to reduce the complexity of the data, making the subsequent data analysis and model construction more efficient and accurate, and providing important basis and support for in-depth research on the impact mechanism of air pollutants on respiratory diseases and the formulation of relevant environmental protection and health strategies;

[0048] (2) In order to solve the technical problem that the spatial information of air pollutant emission data is incomplete and it is difficult to extract valuable air pollutant emission data, multi-scale information fusion technology is used to fuse multi-scale information particles, and the significance of each scale is calculated based on the difference in data distribution to better grasp the data characteristics;

[0049] (3) In order to address the technical problem that respiratory disease predictions have large errors, misleading the public health department's judgment on disease epidemic trends, thereby affecting the formulation of effective prevention and control strategies and measures, the ABC algorithm is used to optimize the tree parameters and construct a CART tree. The tree structure is precisely controlled to achieve effective tree pruning and node setting to ensure that the prediction error is minimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of a respiratory disease prediction system based on machine learning provided by the present invention;

[0051] Figure 2 It is a schematic diagram of the data preprocessing module;

[0052] Figure 3 is a schematic diagram of the feature extraction module;

[0053] Figure 4 Schematic diagram of the module for building a respiratory disease prediction model.

[0054] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0056] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0057] Example 1, see Figure 1 , a respiratory disease prediction system based on machine learning provided by the present invention includes a data preparation module, a data preprocessing module, a feature extraction module, a respiratory disease prediction model building module, an analysis model output module and a report generation module;

[0058] The data preparation module specifically collects data on air pollutant emissions and the number of patients with respiratory diseases;

[0059] The data preprocessing module specifically adopts an unsupervised discretization technology based on an artificial neural network self-organizing map, trains the network through a competitive learning mechanism to distribute the air pollutant emission data to learn potential patterns and features, and maps and discretizes the air pollutant emission data according to clustering areas;

[0060] The feature extraction module specifically uses a multi-label and multi-scale learning strategy to mine potential multiple classification information, uses correlation differences to fit true labels, uses distribution differences to correct ratio deviations, builds a model for missing label multi-label data, and calculates significance;

[0061] The respiratory disease prediction model building module specifically builds a CART tree, uses the ABC algorithm to prune leaves and set split points, calculates the fitness unit of the solution vector, uses the training error as the fitness function, calculates the fitness of the solution vector, and ensures that the learning model prediction error is minimized;

[0062] The analysis model output module is specifically used to visualize data and display the model's prediction results of respiratory diseases in the form of intuitive charts;

[0063] The report generation module is specifically for presenting and explaining the prediction results in detail so that non-professionals can also effectively understand it and summarize the model as a whole.

[0064] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the data preparation module, the data preparation module specifically collects long-term series of air pollutant emission data and the number of patients with respiratory diseases in the same period. The air pollutants include PM2.5, PM10, sulfur dioxide and nitrogen oxides.

[0065] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In the data preprocessing module, a data cleaning unit and a data analysis processing unit are provided. The data preprocessing module includes the following contents:

[0066] Data cleaning unit, which collects data on air pollutant emissions and the number of patients with respiratory diseases, removes noise, corrects erroneous data, and handles missing values;

[0067] The data analysis and processing unit uses unsupervised discretization technology based on artificial neural network self-organizing map to analyze and process air pollutant emission data, including the following:

[0068] Network initialization: determine the number of nodes and topological structure of the self-organizing graph network, and randomly initialize the node vector;

[0069] Training the network, inputting air pollutant emission data into the network, and using a competitive learning mechanism to gradually adapt the nodes to the distribution of air pollutant emission data, so as to learn the potential patterns and features in the air pollutant emission data;

[0070] Forming clusters, after training, the network nodes will form different cluster areas to reflect the similarity and correlation between air pollutant emission data;

[0071] Discrete mapping maps and discretizes the air pollutant emission data according to the clustering area to which it belongs in the network, and determines the discrete category corresponding to each air pollutant emission data. The discrete categories include discretization according to the concentration range of the pollutant and discretization according to the type of pollutant.

[0072] By performing the above operations, the complexity of the data is reduced by discretizing the data, making the subsequent data analysis and model building more efficient and accurate, providing important basis and support for in-depth research on the impact mechanism of air pollutants on respiratory diseases and formulating relevant environmental protection and health strategies, and solving the technical problems of difficulty in finding potential correlation patterns between air pollutants and patients with respiratory diseases, and complex original data that are difficult to directly analyze.

[0073] Example 4, see Figure 1and Figure 3 This embodiment is based on the above embodiment. In the feature extraction module, a spatial conversion unit, a multi-scale information particle fusion unit, a true label fitting unit, a significance calculation unit of each scale and a feature classification unit are provided. The feature extraction module includes the following contents:

[0074] The spatial transformation unit transforms a single data space into a multi-scale fuzzy granular space through a fuzzy neighborhood granularity recognition mechanism, and uses a multi-label multi-scale learning strategy to mine potential multi-classification information;

[0075] The multi-scale information particle units are integrated to construct a multi-scale fusion fuzzy granularity space to ensure the diversity and integrity of pollutant information in the air pollutant emission data space, and the value of the multi-label multi-scale fuzzy dependency function at the k scale is calculated. The formula used is as follows:

[0076] ;

[0077] In the formula, represents the value of the multi-label multi-scale fuzzy dependency function at scale k, B represents the air pollutant emission category dataset, U represents the full set of data, Represents the i-th element in the data set, i represents the index of the element in the data set, It represents the correlation degree between the element and the air pollutant emission category dataset B at the k scale, Represents a fuzzy label set and reconstructs a new label set based on the data with missing labels;

[0078] Fitting the true label unit follows the principle that samples with greater correlation in the multi-scale fusion granularity space have more similar labels, and uses the difference in the correlation ratio between labeled and unlabeled samples in the sample granularity structure to fit the true label;

[0079] Calculate the significance units of each scale, use the distribution difference in the data to correct the ratio deviation, and build a multi-scale fuzzy rough set model for multi-label data with missing labels to calculate the significance of each scale. The formula used is as follows:

[0080] ;

[0081] In the formula, represents the significance at scale k, represents the data distribution difference at this scale, and f represents the functional relationship for calculating the significance based on the distribution difference;

[0082] The feature classification unit defines the multi-scale fusion fuzzy uncertainty metric, explores the correlation between features and fuzzy label sets, the redundancy and interactivity between features, designs corresponding feature evaluation criteria, selects discriminant features for classification, and thus extracts valuable air pollutant emission data for further analysis.

[0083] By performing the above operations, multi-scale information fusion technology is used to fuse multi-scale information particles, and the significance of each scale is calculated using data distribution differences to better grasp data characteristics, thus solving the technical problem of incomplete spatial information of air pollutant emission data and difficulty in extracting valuable air pollutant emission data.

[0084] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In the module for constructing a respiratory disease prediction model, an ABC algorithm optimization tree unit, a CART construction unit, an optimization variable unit, and a fitness unit for calculating a solution vector are provided. The module for constructing a respiratory disease prediction model includes the following contents:

[0085] The ABC algorithm optimizes the tree unit and uses the ABC algorithm to optimize and prune the parameters of the tree. The tree components include decision nodes and leaf nodes. Each decision node represents one of the constraints that constitute the tree rules. Based on the constraint, the input sample is analyzed according to the condition of one of the decision variables, and the fitness function of the ETC-related indicators is calculated. The formula used is as follows:

[0086] ;

[0087] In the formula, represents the fitness function of ETC-related indicators, T represents the specified target vector, C represents the function for calculating the correlation coefficient, N1 represents the number of samples, , and Represents a specific sample parameter, i, j and a are the indexes of the sample;

[0088] CART construction unit,The CART construction process consists of selecting input variables and,determining the split points of these variables until a suitable tree is created,which is done using a greedy algorithm to minimize the cost function.,In each splitting step, the point with the lowest cost is selected;

[0089] The CART model is used for classification, and the Gini impurity index is used to determine the cost function for building the tree. The formula used is as follows:

[0090] ;

[0091] In the formula, Gini represents Gini impurity, which is used to measure the degree of confusion of the data set, C represents the number of target classes, that is, the total number of different categories to be classified, and p c1 Indicates the proportion of training instances marked as category c1 in the current node;

[0092] Optimize variable units. After building the initial CART model, use the ABC algorithm to prune leaves and set split points. For a decision tree with N decision nodes and M leaf nodes, optimize variables, including the following:

[0093] N real variables, each variable corresponds to a node of the decision tree, and the value of the variable represents the split point of the node. Since all input variables of the problem are normalized using the formula, the search boundaries of all variables are (0, 1);

[0094] M binary variables, each binary variable corresponds to a leaf node of the tree. In the binary variables, a value of zero indicates pruning the branch corresponding to the leaf node in the tree structure, and a value of one indicates the maintenance of the node in the tree structure;

[0095] Each solution vector in the ABC algorithm has a heterogeneous structure. In each solution vector, the first N elements are used to determine the split point of the decision node, and the last M elements specify the deletion and retention of leaf nodes in the tree structure;

[0096] The fitness unit of the solution vector is calculated. Each solution vector is applied to the initial tree structure. The fitness of the solution vector is calculated based on the resulting tree. The training error criterion is used as the fitness function in the ABC algorithm. The validation sample is applied to the tree modified by the solution vector, and the output label of the tree is compared with the actual label. The fitness of the solution vector is calculated using the formula. The formula used is as follows:

[0097] ;

[0098] Where Fit represents the fitness of the solution vector, E represents the number of validation instances where the output of the tree is different from the actual label of the instance, and N represents the total number of real variables. By using the proposed structure for the solution vector, the decision nodes in the learning model are effectively set while pruning the tree. The training error criterion is used as the fitness function in the optimization algorithm to ensure that the prediction error in the learning model is minimized.

[0099] By performing the above operations, the ABC algorithm is used to optimize the tree parameters and prune them, a CART tree is constructed, and the tree structure is precisely regulated to achieve effective tree pruning and node setting, ensuring that the prediction error is minimized. This solves the technical problem that respiratory disease predictions have large errors, misleading the public health department's judgment on disease epidemic trends, and thus affecting the formulation of effective prevention and control strategies and measures.

[0100] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the analysis model output module, data visualization is performed specifically to display the model's prediction results for respiratory diseases in the form of intuitive charts, including a bar chart of the prediction ratios of different disease categories and a trend chart of individual patient prediction results, and analyzes the different changing trends and rules of the prediction results under different air pollutant emission data.

[0101] Embodiment 7, see Figure 1 ,This embodiment is based on the above embodiment, and specifically presents the prediction results in the report generation module, including the values ​​of various indicators, chart displays and case analysis, and explains the prediction results so that non-professionals can also better understand them, and summarizes the overall performance and effectiveness of the model and its value for predicting respiratory diseases.

[0102] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0103] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0104] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A respiratory disease prediction system based on machine learning, characterized by: The system includes a data preparation module, a data preprocessing module, a feature extraction module, a respiratory disease prediction model building module, an analysis model output module and a report generation module; The data preparation module specifically collects data on air pollutant emissions and the number of patients with respiratory diseases; The data preprocessing module specifically adopts an unsupervised discretization technology based on an artificial neural network self-organizing map, trains the network through a competitive learning mechanism to distribute the air pollutant emission data to learn potential patterns and features, and maps and discretizes the air pollutant emission data according to clustering areas; The feature extraction module specifically uses a multi-label and multi-scale learning strategy to mine potential multiple classification information, uses correlation differences to fit true labels, uses distribution differences to correct ratio deviations, builds a model for missing label multi-label data, and calculates significance; The respiratory disease prediction model building module specifically builds a CART tree, uses the ABC algorithm to prune leaves and set split points, calculates the fitness unit of the solution vector, uses the training error as the fitness function, calculates the fitness of the solution vector, and ensures that the learning model prediction error is minimized; The analysis model output module is specifically used to visualize data and display the model's prediction results of respiratory diseases in the form of intuitive charts; The report generation module is specifically for displaying and explaining the prediction results in detail.

2. A respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the feature extraction module, there are provided a space conversion unit, a multi-scale information particle fusion unit, a true label fitting unit, a significance calculation unit of each scale, and a feature classification unit. The feature extraction module includes the following contents: The spatial transformation unit transforms a single data space into a multi-scale fuzzy granular space through a fuzzy neighborhood granularity recognition mechanism, and uses a multi-label multi-scale learning strategy to mine potential multi-classification information; The multi-scale information particle units are integrated to construct a multi-scale fusion fuzzy granularity space to ensure the diversity and integrity of pollutant information in the air pollutant emission data space, and the value of the multi-label multi-scale fuzzy dependency function at the k scale is calculated. The formula used is as follows: ； In the formula, represents the value of the multi-label multi-scale fuzzy dependency function at scale k, B represents the air pollutant emission category dataset, U represents the full set of data, Represents the i-th element in the data set, i represents the index of the element in the data set, It represents the correlation degree between the element and the air pollutant emission category dataset B at the k scale, Represents a fuzzy label set and reconstructs a new label set based on the data with missing labels; Fitting the true label unit follows the principle that samples with greater correlation in the multi-scale fusion granularity space have more similar labels, and uses the difference in the correlation ratio between labeled and unlabeled samples in the sample granularity structure to fit the true label; Calculate the significance units of each scale, use the distribution difference in the data to correct the ratio deviation, and build a multi-scale fuzzy rough set model for multi-label data with missing labels to calculate the significance of each scale. The formula used is as follows: ； In the formula, represents the significance at scale k, represents the data distribution difference at this scale, and f represents the functional relationship for calculating the significance based on the distribution difference; The feature classification unit defines the multi-scale fusion fuzzy uncertainty metric, explores the correlation between features and fuzzy label sets, the redundancy and interactivity between features, designs corresponding feature evaluation criteria, selects discriminant features for classification, and thus extracts valuable air pollutant emission data for further analysis.

3. The respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the module for constructing a respiratory disease prediction model, an ABC algorithm optimization tree unit, a CART construction unit, an optimization variable unit, and a fitness unit for calculating a solution vector are provided. The module for constructing a respiratory disease prediction model includes the following contents: The ABC algorithm optimizes the tree unit, uses the ABC algorithm to optimize and prune the parameters of the tree, and calculates the fitness function of ETC-related indicators. The formula used is as follows: ； In the formula, represents the fitness function of ETC-related indicators, T represents the specified target vector, C represents the function for calculating the correlation coefficient, N1 represents the number of samples, , and Represents a specific sample parameter, i, j and a are the indexes of the sample; CART construction unit,The CART construction process consists of selecting input variables and,determining the split points of these variables until a suitable tree is created,which is done using a greedy algorithm to minimize the cost function.,In each splitting step, the point with the lowest cost is selected; The CART model is used for classification, and the Gini impurity index is used to determine the cost function for building the tree. The formula used is as follows: ； In the formula, Gini represents Gini impurity, which is used to measure the degree of confusion of the data set, C represents the number of target classes, that is, the total number of different categories to be classified, and p c1 Indicates the proportion of training instances marked as category c1 in the current node; Optimize variable units. After building the initial CART model, use the ABC algorithm to prune leaves and set split points. For a decision tree with N decision nodes and M leaf nodes, optimize variables, including the following: N real variables, each variable corresponds to a node of the decision tree, and the value of the variable represents the split point of the node. Since all input variables of the problem are normalized using the formula, the search boundaries of all variables are (0, 1); M binary variables, each binary variable corresponds to a leaf node of the tree. In the binary variables, a value of zero indicates pruning the branch corresponding to the leaf node in the tree structure, and a value of one indicates the maintenance of the node in the tree structure; Each solution vector in the ABC algorithm has a heterogeneous structure. In each solution vector, the first N elements are used to determine the split point of the decision node, and the last M elements specify the deletion and retention of leaf nodes in the tree structure; The fitness unit of the solution vector is calculated. Each solution vector is applied to the initial tree structure. The fitness of the solution vector is calculated based on the resulting tree. The training error criterion is used as the fitness function in the ABC algorithm. The validation sample is applied to the tree modified by the solution vector, and the output label of the tree is compared with the actual label. The fitness of the solution vector is calculated using the formula. The formula used is as follows: ； Where Fit represents the fitness of the solution vector, E represents the number of validation instances where the output of the tree is different from the actual label of the instance, and N represents the total number of real variables.

4. The respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the data preprocessing module, a data cleaning unit and a data analysis processing unit are provided. The data preprocessing module includes the following contents: Data cleaning unit, which collects data on air pollutant emissions and the number of patients with respiratory diseases, removes noise, corrects erroneous data, and handles missing values; The data analysis and processing unit uses unsupervised discretization technology based on artificial neural network self-organizing map to analyze and process air pollutant emission data, including the following: Network initialization: determine the number of nodes and topological structure of the self-organizing graph network, and randomly initialize the node vector; Training the network, inputting air pollutant emission data into the network, and using a competitive learning mechanism to gradually adapt the nodes to the distribution of air pollutant emission data, so as to learn the potential patterns and features in the air pollutant emission data; Forming clusters, after training, the network nodes will form different cluster areas to reflect the similarity and correlation between air pollutant emission data; Discrete mapping maps and discretizes the air pollutant emission data according to the clustering area to which it belongs in the network, and determines the discrete category corresponding to each air pollutant emission data. The discrete categories include discretization according to the concentration range of the pollutant and discretization according to the type of pollutant.

5. The respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the data preparation module, the data preparation module specifically collects long-term series of air pollutant emission data and data on the number of patients with respiratory diseases in the same period.

6. A respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the analysis model output module, data visualization is performed specifically to display the model's prediction results for respiratory diseases in the form of intuitive charts, including a bar chart of the prediction ratio of different disease categories and a trend chart of individual patient prediction results, to analyze the different changing trends and patterns of the prediction results under different air pollutant emission data.

7. The respiratory disease prediction system based on machine learning according to claim 1, characterized in that: In the report generation module, the prediction results are presented specifically, including the numerical values ​​of various indicators, chart displays and case analyses. The prediction results are displayed and explained so that non-professionals can understand them, summarizing the overall performance and effectiveness of the model and its value in predicting respiratory diseases.

Images