Occupational health examination result prediction method based on expert system and machine learning

Abstract

The invention discloses an occupational health examination result prediction method based on an expert system and machine learning, and the method comprises the steps: obtaining physical examination archive information, carrying out the cleaning and standardization, adding doctor seeing frequency information to form an original data set, extracting important features through deep learning, and carrying out the recombination to obtain a new data set; identifying all hazard factor types of each physical examination person in the new data set; performing logic judgment on each hazard factor according to an occupational health standard to obtain a corresponding prediction label so as to form a decision set; respectively marking occupational contraindications and suspected occupational diseases for the decision set, marking the feature SOD and the feature OC corresponding to the result, and adding the feature SOD and the feature OC into a new data set; randomly dividing the new data set passing through the expert system link into a training set and a test set, and selecting more than two machine learning models to perform training by using the training set to obtain respective prediction results; and carrying out weighted average on prediction results of more than two models through a weighted voting method to obtain a final output result. The occupational health risk can be identified more effectively.

Description

Technical Field

[0001] The present invention relates to the field of predictive medicine, and in particular to a method for predicting occupational health examination results based on expert systems and machine learning. Background Art

[0002] The basic task of predictive medicine is to predict the probability of an individual contracting a certain disease, which is crucial for developing effective preventive measures, aiming to completely prevent the disease or at least reduce the impact of the disease on the patient. In recent years, data mining and machine learning techniques have made significant progress in solving predictive medicine and bioinformatics problems. However, due to the diversity of data sets and the particularity of predictive medicine tasks, there are relatively few studies on data analysis, processing, mining, and predictive models for the diagnostic conclusions of occupational health examination results. Although there have been some analytical processing and mining works in other related data sets, these studies often do not explore the characteristics of the data in depth, and the models used are not ideal in terms of prediction effect, and there are limitations in practical applications.

[0003] The performance of the existing k-nearest neighbor (KNN) algorithm depends largely on the quality and quantity of training data. If there is bias or incompleteness in the training data, the prediction accuracy of the model will be affected. Secondly, when processing large-scale data sets, the KNN algorithm needs to calculate the distance between the sample to be predicted and all training samples, so the computational complexity is high, resulting in a slow prediction process. In addition, the KNN algorithm is highly sensitive to feature scale, so feature normalization is usually required before application.

[0004] The support vector machine (SVM) algorithm is very sensitive to parameter selection, especially the regularization parameter C and the kernel function parameter. The selection of these parameters usually needs to be optimized through methods such as cross-validation. Secondly, SVM may encounter computational efficiency issues when processing large-scale data sets, especially when the feature dimension is very high. In addition, the interpretability of the SVM model is relatively poor, the so-called "black box" problem, which may be a challenge in the field of occupational health examination and diagnosis where model interpretability is required.

[0005] In addition, although the random forest (RF) algorithm has shown significant advantages in the prediction of occupational health examination and occupational disease diagnosis conclusions, it also faces some challenges and limitations in practical applications. First, due to the complexity of the random forest model, it may contain a large number of decision trees, which not only increases the training time of the model, but also may lead to high memory consumption. In addition, the interpretability of the model is also affected to a certain extent, because the prediction results of many decision trees are integrated, making the decision process of the model less transparent and difficult to explain to non-professional users. The performance of random forests depends largely on the quality and representativeness of the training data, and the bias or incompleteness of the data may affect the prediction accuracy of the model. At the same time, although random forests reduce the risk of overfitting through ensemble methods, overfitting may still occur if the model parameters, such as the number of trees and the number of feature selections, are not adjusted properly. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting occupational health examination results based on expert system and machine learning.

[0007] The technical solution adopted by the present invention is:

[0008] The method for predicting occupational health examination results based on expert system and machine learning includes the following steps:

[0009] Step 1, data processing: obtain the physical examination file information, clean and standardize it, and then add the number of visits information to form the original data set; use the neural network model to perform deep learning on some feature data of the original data set to extract important features; the important features and the remaining feature data of the original data set together form a new feature data set;

[0010] Furthermore, step 1 specifically includes the following steps:

[0011] Step 1-1, traverse the physical examination file information and delete the redundant feature information that is not related to the physical examination conclusion;

[0012] Furthermore, the redundant characteristic information is personal information characteristics, and the personal information characteristics include name, physical examination number, employer name, and employer region.

[0013] Step 1-2, extracting hazard factor information from the physical examination file information; converting each single hazard factor in the hazard factor information into a corresponding dummy variable to form a dummy variable set, and mapping the single hazard factor name of each dummy variable to the hazard factor name classified in the specified occupational health standard, namely the "Technical Specifications for Occupational Health Monitoring"; wherein the hazard factor information includes physical examination hazard factors and contact hazard factors;

[0014] Specifically, the two features of physical examination hazard factors and contact hazard factors are obtained, and a single hazard factor is identified with a comma as a separator and converted into a dummy variable F. A dictionary is established to map the single hazard factor name of F to the hazard factor name classified in the occupational health standard, namely the Technical Specifications for Occupational Health Monitoring, and the two features of physical examination hazard factors and contact hazard factors are deleted and the dummy variable F is added.

[0015] Step 1-3, identify and obtain time feature information from the physical examination file information, convert the months uniformly and save them in integer form;

[0016] Step 1-4, add the number of visits information based on the extracted information of the physical examination file information to form the original data set;

[0017] Steps 1-5, construct a neural network model, input part of the feature data of the original data set into the neural network model to perform deep learning on the original data set to extract important features; the important features and the remaining feature data of the original data set together form a new feature data set.

[0018] Furthermore, the time feature information in steps 1-3 includes total length of service, length of service with injuries, and start and end dates.

[0019] Furthermore, the neural network model in steps 1-5 includes an input layer, a hidden layer and an output layer; the input layer is used to receive data with a dimension equal to the number of features in the training set; the hidden layer includes two linear layers, the first linear layer maps the input features to a 256-dimensional feature vector, and Dropout regularization is added after the first linear layer, and the dropout rate of Dropout regularization is 0.2; the second linear layer compresses the 256-dimensional feature vector outputted by the first linear layer to obtain a 128-dimensional feature vector; each linear layer is followed by a ReLU activation function to introduce nonlinearity and enhance the expressive power of the model; the linear layer of the output layer maps the 128-dimensional feature vector to the number of target categories, i.e., 5 dimensions.

[0020] Furthermore, the neural network model in steps 1-5 uses the cross entropy loss function and the Adam optimizer to train the network.

[0021] Specifically, the neural network model performs the following actions: the input layer receives data with dimensions equal to the number of features in the training set; then, the data is feature mapped through two linear layers. The first linear layer maps the data to 256 dimensions, and the second linear layer further compresses it to 128 dimensions. After each linear layer, the ReLU activation function is applied to increase the nonlinear ability of the model. Dropout regularization is added after the first linear layer, and its dropout rate is 0.2. Finally, the 128-dimensional feature vector is mapped to 5 dimensions through a linear layer to complete the forward propagation process of the entire network. The network is trained using the cross entropy loss function and the Adam optimizer during the training process. After the model training is completed, the trained network is used to extract features from new data. Finally, a new data set is obtained by combining the second linear layer data features extracted by the neural network and other features in the original data set.

[0022] Specifically, the second linear layer data features extracted by the neural network represent new features extracted by the trained neural network that affect the prediction results, which can be understood as new latent features discovered by the association between certain key features in the data set.

[0023] Step 2: Identify all types of hazard factors for each physical examination personnel in the new data set; perform logical judgment on each hazard factor according to the provisions of the specified occupational health standard (specifically, the National Occupational Health Standard of the People's Republic of China GBZ188-2014 "Technical Specifications for Occupational Health Monitoring") to obtain the corresponding prediction label, and summarize it to form a decision set; mark the occupational contraindications and suspected occupational diseases in the decision set, and add the features SOD and OC corresponding to the marked results to the new data set;

[0024] Furthermore, the prediction labels include “no abnormality found at present”, “recheck”, “suspected occupational disease”, “occupational contraindications” and other diseases or abnormalities.

[0025] Furthermore, the specific steps of step 2 are as follows:

[0026] Step 2-1, identify all types of hazard factors of the physical examination personnel in the data set: F = {f1,f2,...,f n}, n represents the number of single hazard factor features; for each physical examination record, the single hazard factor feature set F is traversed to obtain the single hazard factor feature set G corresponding to each physical examination record. The feature set G of the i-th physical examination record is G i ; G = {g1, g2, …, g k}, k is the number of single hazard factor characteristics with a value equal to 1;

[0027] Specifically, as a feasible implementation, the number of single hazard factor features n=81; the single hazard factor features with a value of 1 in the single hazard factor feature set F are obtained to form a single hazard factor feature set G of each physical examination record.

[0028] Step 2-2, identify and obtain the physical examination feature t corresponding to the hazard factor f of each physical examination record, where T is the set of physical examination features t; the physical examination feature set T of the i-th physical examination record is T i :T={t1,t3,…,t d ,d is the number of physical examination features};

[0029] Step 2-3, construct a logic function H for each hazard factor f. The specific expression is as follows:

[0030]

[0031] Step 2-4: Establish a corresponding decision set D for a single hazard factor feature set G, D = {d1, d2, ..., d k ,d i =H(f i ,T i )},d i The single hazard factor g represents the corresponding single hazard factor feature set G i The value judged under the logic function H; the physical examination feature set D of the jth physical examination record is D j :

[0032] Step 2-5, mark the occupational contraindications and suspected occupational diseases in the decision set to obtain new features SOD and OC, respectively, and add the new features SOD and OC to the new data set;

[0033] Among them, the feature SOD represents the expert system simulating the expert decision-making on suspected occupational diseases. 1 indicates the existence of suspected occupational diseases, and 0 indicates the absence of such diseases. The feature OC represents the judgment of occupational contraindications by the expert system simulating expert decision-making. 1 means the occupational contraindication exists, and 0 means the contraindication does not exist.

[0034] Step 3, hybrid model training: randomly divide the new data set that has passed the expert system link into a training set and a test set, and select two or more machine learning models to use the training set for training to obtain their respective prediction results; use the weighted voting method to weight the prediction results of the two or more models to obtain the final output result.

[0035] Furthermore, four models including Random Forest, AdaBoost, XGBoost and LightGBM were selected.

[0036] Furthermore, the accuracy of the test set in each learning model is used as the evaluation indicator, and grid search is used to find the hyperparameters of each learning model.

[0037] Specifically, taking accuracy as the evaluation indicator, grid search is used to find the hyperparameters of Random Forest, AdaBoost, XGBoost and LightGBM models respectively. Through the weighted voting method, the prediction results of the four models are weighted averaged to obtain the final output result.

[0038] Furthermore, the specific steps of step 3 are as follows:

[0039] Step 3-1, randomly divide the new data set after the expert system into a training set and a test set in a ratio of 7:3; use the training set to train two or more machine learning models respectively, use grid search to set the hyperparameter range of each learning model, and use the accuracy of the test set prediction on the model as an evaluation indicator to find the hyperparameters of the model;

[0040] Specifically, the selected machine learning models are Random Forest, AdaBoost, XGBoost and LightGBM models.

[0041] Step 3-2: The prediction results of two or more models are weighted averaged to obtain the final output result by weighted voting, in which the accuracy of each model in the test set is used as the evaluation index to assign the weight of each model. The formula for weighted voting is as follows:

[0042]

[0043] in, is the prediction result of the i-th model, ω i is the weight of the i-th model, and the weight is assigned to each model according to its performance on the test set.

[0044] The present invention adopts the above technical scheme, proposes an occupational health examination result diagnosis conclusion analysis and prediction model that integrates machine learning model and expert system, improves the intelligent level of occupational health management, and provides strong support for the prevention and control of occupational diseases. Through this innovative model, occupational health risks can be more effectively identified, providing a scientific basis for the formulation of preventive measures and health policies. Compared with traditional models based on k-nearest neighbors, support vector machines, and random forests, the present invention improves the model's recognition ability for small samples, and also significantly improves the negative prediction value (NPV), ensuring the accuracy of predictions for key categories. The adaptability and generalization ability of the model to unknown situations are enhanced, ensuring the effectiveness and reliability of the model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments;

[0046] Figure 1 It is a flow chart of a method for predicting occupational health examination results based on expert system and machine learning of the present invention;

[0047] Figure 2 It is a schematic diagram of the flow of simulating expert decision-making by the expert system of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0049] The present invention explores an occupational health examination result diagnosis conclusion analysis prediction model that integrates a machine learning model and an expert system. XGBoost, as an efficient machine learning algorithm, performs well in processing unbalanced data and improving prediction accuracy. It can effectively process various complex data patterns by constructing and optimizing multiple decision trees. Expert systems, especially those based on knowledge reasoning, have been widely used in the field of medical health due to their advantages in simulating expert decision-making processes and providing explanatory reasoning capabilities. The prediction model constructed by integrating XGBoost and expert system technology can predict occupational health examination results more accurately and efficiently. The role of machine learning in the model is to identify key patterns in the data through its powerful classification ability, while the expert system provides a reasoning mechanism based on domain knowledge, so that the model can not only make predictions, but also provide explanations for the predictions. This fusion method combines the data processing capabilities of machine learning and the decision support capabilities of the expert system, so that the prediction model can provide more reliable and transparent prediction results when facing complex occupational health data. In addition, while providing predictions, the expert system can also provide users with possible diagnostic paths and treatment suggestions based on its internal knowledge base, which is very valuable for decision makers in occupational health management. The introduction of expert systems also helps to improve the interpretability of the model, allowing non-professionals to understand the model's prediction basis, thereby enhancing trust in the model's prediction results.

[0050] like Figure 1 As shown in or 2, the present invention discloses a method for predicting occupational health examination results based on expert system and machine learning, which comprises the following steps:

[0051] The following steps are involved:

[0052] Step 1, data processing: obtain the physical examination file information, clean and standardize it, and then add the number of visits information to form the original data set; use the neural network model to perform deep learning on some feature data of the original data set to extract important features; the important features and the remaining feature data of the original data set together form a new feature data set;

[0053] Furthermore, step 1 specifically includes the following steps:

[0054] Step 1-1, traverse the physical examination file information and delete the redundant feature information that is not related to the physical examination conclusion;

[0055] Furthermore, the redundant characteristic information is personal information characteristics, and the personal information characteristics include name, physical examination number, employer name, and employer region.

[0056] Step 1-2, extracting hazard factor information from the physical examination file information; converting each single hazard factor in the hazard factor information into a corresponding dummy variable to form a dummy variable set, and mapping the single hazard factor name of each dummy variable to the hazard factor name classified in the specified occupational health standard, namely the "Technical Specifications for Occupational Health Monitoring"; wherein the hazard factor information includes physical examination hazard factors and contact hazard factors;

[0057] Specifically, the two features of physical examination hazard factors and contact hazard factors are obtained, and a single hazard factor is identified with a comma as a separator and converted into a dummy variable F. A dictionary is established to map the single hazard factor name of F to the hazard factor name classified in the specified occupational health standard, namely the Technical Specifications for Occupational Health Monitoring, and the two features of physical examination hazard factors and contact hazard factors are deleted and the dummy variable F is added.

[0058] Step 1-3, identify and obtain time feature information from the physical examination file information, convert the months uniformly and save them in integer form;

[0059] Step 1-4, add the number of visits information based on the extracted information of the physical examination file information to form the original data set;

[0060] Steps 1-5, construct a neural network model, input part of the feature data of the original data set into the neural network model to perform deep learning on the original data set to extract important features; the important features and the remaining feature data of the original data set together form a new feature data set.

[0061] Furthermore, the time feature information in steps 1-3 includes total length of service, length of service with injuries, and start and end dates.

[0062] Furthermore, the neural network model in steps 1-5 includes an input layer, a hidden layer and an output layer; the input layer is used to receive data with a dimension equal to the number of features in the training set; the hidden layer includes two linear layers, the first linear layer maps the input features to a 256-dimensional feature vector, and Dropout regularization is added after the first linear layer, and the dropout rate of Dropout regularization is 0.2; the second linear layer compresses the 256-dimensional feature vector outputted by the first linear layer to obtain a 128-dimensional feature vector; each linear layer is followed by a ReLU activation function to introduce nonlinearity and enhance the expressive power of the model; the linear layer of the output layer maps the 128-dimensional feature vector to the number of target categories, i.e., 5 dimensions.

[0063] Furthermore, the neural network model in steps 1-5 uses the cross entropy loss function and the Adam optimizer to train the network.

[0064] Specifically, the neural network model performs the following actions: the input layer receives data with dimensions equal to the number of features in the training set; then, the data is feature mapped through two linear layers. The first linear layer maps the data to 256 dimensions, and the second linear layer further compresses it to 128 dimensions. After each linear layer, the ReLU activation function is applied to increase the nonlinear ability of the model. Dropout regularization is added after the first linear layer, and its dropout rate is 0.2. Finally, the 128-dimensional feature vector is mapped to 5 dimensions through a linear layer to complete the forward propagation process of the entire network. The network is trained using the cross entropy loss function and the Adam optimizer during the training process. After the model training is completed, the trained network is used to extract features from new data. Finally, a new data set is obtained by combining the second linear layer data features extracted by the neural network and other features in the original data set. Specifically, the second linear layer data features extracted by the neural network represent the new features that affect the prediction results of the trained neural network, which can be understood as new hidden features discovered by the association between certain key features in the data set.

[0065] Step 2: Identify all types of hazard factors for each physical examination personnel in the new data set; perform logical judgment on each hazard factor according to the provisions of the specified occupational health standard (specifically, the National Occupational Health Standard of the People's Republic of China GBZ 188-2014 "Technical Specifications for Occupational Health Monitoring") to obtain the corresponding prediction label, and summarize it to form a decision set; mark the occupational contraindications and suspected occupational diseases in the decision set, and add the features SOD and OC corresponding to the marked results to the new data set;

[0066] Furthermore, the prediction labels include “no abnormality found at present”, “recheck”, “suspected occupational disease”, “occupational contraindications” and other diseases or abnormalities.

[0067] Furthermore, the specific steps of step 2 are as follows:

[0068] Step 2-1, identify all types of hazard factors of the physical examination personnel in the data set: F = {f1,f2,...,f n}, n represents the number of single hazard factor features; for each physical examination record, the single hazard factor feature set F is traversed to obtain the single hazard factor feature set G corresponding to each physical examination record. The feature set G of the i-th physical examination record is G i ; G = {g1, g2, …, g k}, k is the number of single hazard factor characteristics with a value equal to 1;

[0069] Specifically, as a feasible implementation, the number of single hazard factor features n=81; the single hazard factor features with a value of 1 in the single hazard factor feature set F are obtained to form a single hazard factor feature set G of each physical examination record.

[0070] Step 2-2, identify and obtain the physical examination feature t corresponding to the hazard factor f of each physical examination record, where T is the set of physical examination features t; the physical examination feature set T of the i-th physical examination record is T i :T={t1,t3,…,t d ,d is the number of physical examination features};

[0071] Step 2-3, construct a logic function H for each hazard factor f. The specific expression is as follows:

[0072]

[0073] Step 2-4: Establish a corresponding decision set D for a single hazard factor feature set G, D = {d1, d2, ..., d k ,d i =H(f i ,T i )},d i The single hazard factor g represents the corresponding single hazard factor feature set G i The value judged under the logic function H; the physical examination feature set D of the jth physical examination record is D j :

[0074] Step 2-5, mark the occupational contraindications and suspected occupational diseases in the decision set to obtain new features SOD and OC, respectively, and add the new features SOD and OC to the new data set;

[0075] Among them, the feature SOD represents the expert system simulating the expert decision-making on suspected occupational diseases. 1 indicates the existence of suspected occupational diseases, and 0 indicates the absence of such diseases. The feature OC represents the judgment of occupational contraindications by the expert system simulating expert decision-making. 1 means the occupational contraindication exists, and 0 means the contraindication does not exist.

[0076] Step 3, hybrid model training: randomly divide the new data set that has passed the expert system link into a training set and a test set, and select two or more machine learning models to use the training set for training to obtain their respective prediction results; use the weighted voting method to weight the prediction results of the two or more models to obtain the final output result.

[0077] Furthermore, four models including Random Forest, AdaBoost, XGBoost and LightGBM were selected.

[0078] Furthermore, the accuracy of the test set in each learning model is used as the evaluation indicator, and grid search is used to find the hyperparameters of each learning model.

[0079] Specifically, taking accuracy as the evaluation indicator, grid search is used to find the hyperparameters of Random Forest, AdaBoost, XGBoost and LightGBM models respectively. Through the weighted voting method, the prediction results of the four models are weighted averaged to obtain the final output result.

[0080] Furthermore, the specific steps of step 3 are as follows:

[0081] Step 3-1, randomly divide the new data set after the expert system into a training set and a test set in a ratio of 7:3; use the training set to train two or more machine learning models respectively, use grid search to set the hyperparameter range of each learning model, and use the accuracy of the test set prediction on the model as an evaluation indicator to find the hyperparameters of the model;

[0082] Specifically, the selected machine learning models are Random Forest, AdaBoost, XGBoost and LightGBM models.

[0083] In step 3-2, the prediction results of two or more models are weighted averaged to obtain the final output result through the weighted voting method, in which the accuracy of each model in the test set is used as the evaluation index to assign the weight to each model.

[0084] The principle of the present invention is described in detail below:

[0085] The dataset used in this invention was obtained from a coastal capital city in China. The dataset contains 71,084 records and 1,535 features. The dataset is the physical examination file information of the physical examination personnel of the employers in Fuzhou during their on-the-job occupational health examinations in 2021-2022. The physical examination personnel work in different scenarios under different occupational hazard factors. Occupational hazard factors refer to factors or conditions that are generated and (or) exist in occupational activities and may have adverse effects on the health, safety and working ability of occupational groups, including chemical, physical, biological and other factors. Therefore, it is necessary to conduct different physical examination items according to different occupational hazard factors (represented as features in the dataset), and draw corresponding physical examination conclusions according to the "Technical Specifications for Occupational Health Monitoring" (GBZ188).

[0086] Based on the results of occupational health examinations, the physical examination conclusions of the examinees can be divided into the following five categories: (1) No abnormalities found at present: All examination indicators of this occupational health examination are within the normal range; (2) Reexamination: If one or more abnormalities related to the target disease are found during the examination and need to be reexamined, the content and time of the reexamination should be clearly stated; (3) Suspected occupational disease: If the examination finds suspected occupational disease or possible occupational disease, it is necessary to submit it to the occupational disease diagnosis agency for further diagnosis; (4) Occupational contraindications: If the examination finds that the patient has occupational contraindications, the specific disease name should be stated; (5) Other diseases or abnormalities: Other diseases other than the target disease or abnormalities of certain examination indicators. The number and proportion of records of five types of conclusions in this dataset are: other diseases or abnormalities (34,300 records, accounting for 48.25%), no abnormalities found at present (26,950 records, accounting for 37.91%), re-examination (9,273 records, accounting for 13.0%), occupational contraindications (536 records, accounting for 0.75%), and suspected occupational diseases (25 records, accounting for 0.03%).

[0087] Data processing: First, the collected original data set was comprehensively reviewed and evaluated. This data set covers a variety of physical examination results and related individual characteristic information, and aims to analyze the relationship between physical examination results and potential health hazards through machine learning models. However, due to the complexity of the data source, there are many forms of abnormal and redundant information in the original data set. If this information is not properly processed, it will interfere with the subsequent analysis work and affect the accuracy and reliability of the model.

[0088] In order to construct a high-quality data set, the present invention adopts a series of data cleaning and normalization measures. First, I also remove personal information features that are not directly related to the physical examination conclusion, such as name, physical examination number, etc., to protect personal privacy and reduce data redundancy.

[0089] In the processing of time data and character data, the present invention performs unified normalization operations. For time data, the present invention converts the date, year, etc. into a unified format, and extracts the parts useful for analysis (such as month, year, etc.), which are expressed as numerical data to facilitate subsequent calculation and analysis. For character data, the present invention classifies and encodes the data according to its actual meaning, for example, converting the descriptive text of the physical examination results into concise classification labels such as "no abnormality" and "abnormality" so that the model can recognize and judge.

[0090] In the actual process of diagnosing the results of occupational health examinations, different judgments may be drawn for the same physical examination data of the same examinee due to subjective differences among diagnosing doctors. The reason for this difference is that the boundaries between "no abnormality at present", "re-examination", "suspected occupational disease", "occupational contraindications" and other diseases or abnormalities in the prediction labels are not always clear, and the diagnostic tendency and experience of the diagnosing doctors will also affect their judgment. Usually, when the indicators of the physical examination items are abnormal, the doctor needs to comprehensively consider whether the abnormality clearly points to an occupational disease, or is more likely to be a manifestation of other diseases or abnormalities. In the case of uncertainty in diagnosis, doctors often choose the diagnosis of "re-examination" to obtain more information. Therefore, during the first physical examination, doctors may be more inclined to make a re-examination suggestion. In order to capture the dynamic changes in this diagnostic process, the present invention introduces the feature of "number of visits" in the data set to reflect the re-examination of the examinee during the diagnostic process and the evolution of the diagnosis. The addition of this feature helps the model better understand the complexity of diagnostic decisions and improve the accuracy of predicting diagnostic conclusions. In this way, the model can learn the decision-making patterns of different diagnostic doctors when facing similar situations, thereby providing more accurate and consistent prediction results in practical applications.

[0091] In addition, to ensure data quality, the present invention also carefully screened and processed outliers and missing values ​​in the data set. For ambiguous, inconsistent or obviously erroneous data records, the present invention corrected or deleted them; for missing physical examination conclusions or other key information, the present invention adopted corresponding interpolation or deletion strategies based on the severity and distribution of missing data.

[0092] After data cleaning and normalization, the present invention further performs feature engineering to improve the representation ability of the data set and the predictive performance of the model. In particular, for the complex and comprehensive risk factor features in the original data, the present invention splits them into multiple independent single risk factor features. In the original data set, the risk factor may be composed of multiple single risk factors. In order to enable the model to clearly distinguish and evaluate the impact of each single risk factor, the present invention implements dummy variable conversion. In this process, each single risk factor is represented as a dummy variable, where 0 indicates that the risk factor does not exist in the individual, and 1 indicates that it exists. This step not only helps the model to more accurately capture the impact of different risk factors on physical examination results, but also improves the interpretability of the data and the generalization ability of the model.

[0093] In order to better extract the important features in the data set, the present invention adopts deep learning technology, especially neural network, to automatically extract the features of the data set. The present invention designs a feedforward neural network, whose structure includes multiple linear layers and activation layers to achieve effective mapping from raw data to high-dimensional feature space.

[0094] The neural network model consists of the following main parts: Figure 1 As shown:

[0095] Input layer: The input layer of the network receives data of dimension equal to the number of features in the training set, which ensures that the network can process all features.

[0096] Hidden layer: The network contains two linear layers. The first linear layer maps the input features to a 256-dimensional vector, and the second linear layer further compresses these features to 128 dimensions. Each linear layer is followed by a ReLU activation function to introduce nonlinearity and enhance the expressiveness of the model.

[0097] Regularization: To prevent overfitting, the present invention introduces Dropout regularization after the first hidden layer with a dropout rate of 0.2.

[0098] Output layer: Finally, a linear layer maps the 128-dimensional feature vector to the number of target categories, i.e. 5 dimensions.

[0099] The present invention uses a cross entropy loss function and an Adam optimizer to train the network. After the model training is completed, the present invention uses the trained network to extract features from new data. Finally, the present invention obtains a data set containing new features, which combines the features extracted by the neural network with other features in the original data set.

[0100] An expert system is a computer program that simulates the decision-making ability of human experts. It can handle complex decision-making problems and provide expert-level advice or solutions. The core of an expert system is that it can store a large amount of professional knowledge and experience, and use this knowledge to reason and solve problems. In the field of occupational health, an expert system can simulate a doctor's judgment on suspected occupational diseases and occupational contraindications in physical examination records in actual situations.

[0101] The present invention adopts the provisions on hazard factors in the "Technical Specifications for Occupational Health Monitoring" to simulate the doctor's judgment on the two situations of suspected occupational diseases and occupational contraindications recorded in the physical examination in actual situations. The expert system constructs a feature space containing these health indicators by analyzing individual health data, such as lung function test results, biomarker levels in blood samples, long-term occupational exposure history, etc. The system uses this information, combined with professional knowledge and experience, to construct logical functions for logical reasoning, identify individuals with suspected occupational diseases and occupational contraindications, and thus achieve the prediction and diagnosis of occupational diseases.

[0102] (1) Identify all types of hazard factors for physical examination personnel in the data set:

[0103] F={f1,f2,...,f n , n = 81 is the number of single hazard factor characteristics}

[0104] Traverse the single hazard factor feature set F for each physical examination record, and obtain the single hazard factor feature set G with a value of 1. The feature set G of the i-th physical examination record is G i .

[0105] G={g1,g2,…,g k , k is the number of single hazard factor features with a value equal to 1}

[0106] Then, a simulated expert decision is made for each feature of the set G.

[0107] (2) Make a judgment based on the provisions on hazard factors in the National Occupational Health Standard of the People’s Republic of China GBZ 188-2014 “Technical Specifications for Occupational Health Monitoring” in the knowledge base.

[0108] Specific references are: Chapter 5 involves occupational health monitoring of workers exposed to harmful chemical factors, including lead and its inorganic compounds, tetraethyl lead, mercury and its inorganic compounds, manganese and its inorganic compounds, beryllium and its inorganic compounds, cadmium and its inorganic compounds, chromium and its inorganic compounds, zinc oxide, arsenic, arsine (arsine), phosphorus and its inorganic compounds, phosphine, barium compounds (barium chloride, barium nitrate, barium acetate), vanadium and its inorganic compounds, trialkyl tin, thallium and its inorganic compounds, carbonyl nickel, fluorine and its inorganic compounds, benzene (contact with industrial toluene and xylene for reference), carbon disulfide, carbon tetrachloride, methanol, gasoline, methyl bromide, 1,2-dimethylformamide Ethyl chloride, n-hexane, amino and nitro compounds of benzene, trinitrotoluene, benzidine, chlorine, sulfur dioxide, nitrogen oxides, ammonia, phosgene, formaldehyde, monomethylamine, carbon monoxide, hydrogen sulfide, vinyl chloride, trichloroethylene, allyl chloride, chloroprene, organic fluorine, toluene diisocyanate, dimethylformamide, cyanide and nitrile compounds, phenol (phenolic compounds such as cresol, catechol, resorcinol, hydroquinone, etc. shall be implemented as a reference), pentachlorophenol, chloromethyl ether [bis(chloromethyl) ether shall be implemented as a reference], acrylamide, unsymmetrical dimethylhydrazine, dimethyl sulfate, organophosphorus insecticides, carbamate insecticides, pyrethroids, acid mist or acid anhydride, asthma-causing substances, and coke oven emissions. Chapter 6 involves occupational health monitoring of workers in dust-related operations, including free silica dust (crystalline silica dust), coal dust, asbestos dust, other inorganic dusts that cause pneumoconiosis, cotton dust (including flax, soft hemp, jute dust), and organic dust. Chapter 7 involves occupational health monitoring of workers exposed to harmful physical factors, including noise, hand-transmitted vibration, high temperature, high air pressure (see GB 20827), ultraviolet radiation (UV) and microwaves. Chapter 8 involves occupational health monitoring of workers exposed to harmful biological factors, including Brucella and Bacillus anthracis (abbreviated as Bacillus anthracis). Chapter 9 involves occupational health monitoring of workers in special operations, including electrical work, high-altitude work, pressure vessel work, tuberculosis prevention and control work, hepatitis prevention and control work, professional motor vehicle driving work, video work, plateau work, and aviation work.

[0109] According to the provisions of the "Technical Specifications for Occupational Health Monitoring" on hazard factors, the present invention constructs a logical function for each hazard factor. The specific process is as follows:

[0110] Identify and obtain the physical examination features t of the hazard factor f, where T is the set of physical examination features t. The physical examination feature set T of the i-th physical examination record is T i :

[0111] T={t1,t2,…,t d , d is the number of physical examination features}

[0112] Construct a function H for each hazard factor f.

[0113]

[0114] (3) Simulating expert decision-making: Establish a corresponding decision set D for set G, and perform a logical function judgment on each hazard factor f in set G. The physical examination feature set D of the jth physical examination record is D j :

[0115] D={d1,d2,…,d k ,d i =H(f i ,T i )}

[0116] Add two new feature columns to the dataset: SOD (Suspected occupational disease) and OC (Occupational contraindications)

[0117]

[0118] Among them, the feature SOD represents the expert system simulating the expert decision-making judgment on suspected occupational diseases, 1 means the suspected occupational disease exists, and 0 means it does not exist. The feature OC represents the expert system simulating the expert decision-making judgment on occupational contraindications, 1 means the occupational contraindications exist, and 0 means they do not exist.

[0119] A hybrid model based on multiple machine learning algorithms is proposed. By using four classic machine learning models, Random Forest, AdaBoost, XGBoost and LightGBM, an integrated model that can make predictions from multiple dimensions is constructed, and the final output is obtained by weighted averaging the prediction results of each model. In order to ensure the accuracy and generalization ability of the model, the present invention carefully tunes the parameters of each model and adopts a weighted fusion strategy to improve the model performance. The following is a detailed description of the model training process.

[0120] (1) Dataset overview and preprocessing: The dataset used contains 71,084 records, each of which consists of 428 features. Due to its complexity and high dimensionality, it brings two problems: the model is prone to overfitting and the training time is long. Therefore, the present invention performs a series of data cleaning and normalization measures in the data preprocessing stage above. At the same time, the dataset is divided into a training set and a test set.

[0121] (2) Main methods for model hyperparameter training: Grid Search is an exhaustive search method for hyperparameter tuning, which aims to find the hyperparameter combination that optimizes the performance of the machine learning model. Its basic principle is to pre-set the possible value range of each hyperparameter, traverse all possible combinations within these value ranges, train the models one by one, evaluate their performance on the validation set, and finally select the parameter combination with the best performance.

[0122] Specifically, grid search first defines a parameter space, that is, a set of possible values ​​for each hyperparameter. Then, by traversing all possible combinations of these values, the model is trained and the performance is evaluated using a validation set. Each trained model is scored based on an accuracy evaluation metric. Finally, grid search selects the parameter combination that performs best on the validation set as the final hyperparameter setting.

[0123] (3) Model principles and formulas: In order to make full use of the characteristics of different algorithms, the present invention selects four mainstream machine learning models: Random Forest, AdaBoost, XGBoost and LightGBM.

[0124] Random Forest: Random Forest is an ensemble learning method based on decision trees. It improves the overall accuracy of the model by building multiple decision trees and voting on their results. The core idea of ​​random forest is to introduce randomness in the training process: during the construction of each tree, a subset of data and a subset of features are randomly selected to generate multiple independent decision trees. Finally, random forest makes the final prediction by voting or averaging the output results of all decision trees.

[0125] The basic formula of random forest is:

[0126]

[0127] Where N is the number of decision trees, h i (x) is the prediction result of the i-th tree for the input x, and the final output is the average or voting result of the prediction results of all decision trees.

[0128] The main parameters of the random forest model were optimized through grid search. The specific optimization results are as follows:

[0129] 'n_estimators' (number of decision trees): the optimal value is 3000;

[0130] 'max_depth' (maximum depth of the tree): the optimal value is 13;

[0131] 'min_samples_split' (the minimum number of samples required for internal node repartitioning): the optimal value is 2;

[0132] 'min_samples_leaf' (minimum number of samples required for a leaf node): The optimal value is 4.

[0133] After tuning, the accuracy of random forest on the test set is 0.9341, showing strong performance.

[0134] AdaBoost (Adaptive Boosting): AdaBoost (Adaptive Boosting) is a boosting method that aims to combine multiple weak classifiers (such as decision trees) to form a strong classifier. It works by gradually building a model through an iterative process, and each round of the model will focus on the samples that were misclassified in the previous round. In each round of training, the weight of the misclassified samples will increase, and the weight of the correctly classified samples will decrease, so that the next round of weak classifiers will pay more attention to the samples that were misclassified in the previous round.

[0135] The update rule of AdaBoost is as follows:

[0136]

[0137] Among them, ω i is the weight of the i-th sample, h(x i ) is the classifier for sample x i The prediction result, y i is the true label of the sample, α is the weight of the classifier, and Ι represents the relative importance of the classifier.

[0138] The final strong classifier is the weighted result of all weak classifiers:

[0139]

[0140] For the AdaBoost model, the present invention uses the default basic decision tree classifier, adjusts the number of iterations and the learning rate based on grid search, and finally obtains a test set accuracy of 0.9114.

[0141] XGBoost: XGBoost is a reinforcement learning algorithm based on gradient boosting. Compared with the traditional GBDT (Gradient Boosting Decision Tree), XGBoost has been optimized in terms of regularization, missing value processing, parallel processing, etc., thereby improving the performance and efficiency of the model.

[0142] The goal of XGBoost is to minimize the objective function by iteratively building multiple decision trees, where the objective function includes the loss function and the regularization term:

[0143]

[0144] in, is the loss function, which measures the prediction value of the model and the true value y i The error between Ω(f k ) is a regularization term, which is used to control the complexity of the model and prevent overfitting.

[0145] The core idea of ​​XGBoost is to gradually reduce the model error by adding a new decision tree each time to fit the residual of the current model. Specifically, in each iteration, XGBoost updates the model parameters by gradient descent to minimize the sum of squares of the current residual.

[0146] The main parameters of XGBoost are also tuned based on grid search, as follows:

[0147] 'n_estimators' (number of decision trees): the optimal value is 2000;

[0148] 'max_depth' (maximum depth of the tree): the optimal value is 3;

[0149] 'learning_rate': The optimal value is 2.

[0150] The tuned XGBoost model achieved an accuracy of 0.9575 on the test set, making it the best performing model in this experiment.

[0151] LightGBM (Light Gradient Boosting Machine)

[0152] LightGBM (Light Gradient Boosting Machine) is an efficient gradient boosting framework that is particularly suitable for large-scale data sets and high-dimensional features. Similar to XGBoost, LightGBM is also a gradient boosting method based on decision trees, but it is accelerated through histogram optimization (Histogram-based).

[0153] The core idea of ​​LightGBM is to discretize continuous features into several discrete intervals and perform splitting decisions on the histogram, thereby greatly reducing the amount of calculation. Its objective function is similar to XGBoost, and also includes a loss function and a regularization term:

[0154]

[0155] Among them, f(x i ) is the model's prediction for sample x i The predicted value of , l is the loss function, and Ω is the regularization term used to control the complexity of the model.

[0156] Using grid search, the main parameters of LightGBM were tuned as follows:

[0157] 'n_estimators' (number of decision trees): the optimal value is 2000;

[0158] 'max_depth' (maximum depth of the tree): the optimal value is 3;

[0159] 'min_child_samples' (minimum number of samples for leaf nodes): the optimal value is 8;

[0160] 'min_split_gain' (minimum split gain): The optimal value is 0.

[0161] After tuning, the accuracy of the LightGBM model on the test set was 0.9020. Although it performed slightly worse than the four models, its high efficiency in large-scale data processing is worthy of recognition.

[0162] (4) Model fusion and weighted voting: Since each model performs differently under different data features, the present invention adopts a model fusion method to fully utilize the advantages of each model. Through the weighted voting method, the present invention weights and averages the prediction results of the four models to obtain the final output result. The formula for weighted voting is as follows:

[0163]

[0164] in, is the prediction result of the i-th model, ω i is the weight of the model, and the weight is assigned to each model based on its performance on the test set.

[0165] Finally, the prediction results after model fusion are better than those of a single model in terms of accuracy and robustness, with the test accuracy reaching 0.9667, showing the advantages of the hybrid model.

[0166] By building a hybrid model based on Random Forest, AdaBoost, XGBoost and LightGBM, the prediction problem on large-scale high-dimensional data sets was solved. Each model was carefully tuned and optimized, and finally the model was integrated through weighted voting to improve the overall performance. The advantages of combining the various models make the system have high accuracy and stability, and can effectively cope with complex data environments.

[0167] Compared with the prior art, the present invention has the following technical features: (1) The data processing of the present invention effectively solves the complexity, effectiveness and availability of data sources. The present invention extracts the most valuable information from massive data by implementing data cleaning, feature selection and dimensionality reduction technology, and eliminates data with strong interference and low credibility; improves the quality of data and the accuracy of the model, and significantly reduces the computational cost of data processing, making the data mining process more efficient and targeted. (2) The present invention solves the problem of low prediction accuracy of small samples. Since occupational contraindications and suspected occupational diseases account for too small a proportion of the data set, and due to different hazard factors, the physical examination items to be tested are different, and the logical judgment of the diagnosis results is complex, the common mainstream model has poor prediction effect on small sample classification, and even makes the prediction accuracy of occupational contraindications and suspected occupational diseases zero in order to blindly improve the overall accuracy. The present invention introduces an expert system and combines the knowledge and experience of experts in the field to make effective judgments for the two key categories of occupational contraindications and suspected occupational diseases, which not only improves the model's recognition ability for small samples, but also significantly improves the negative prediction value (NPV), ensuring the prediction accuracy of key categories. (3) The present invention solves the practicality of the model in practical application. When the common mainstream model judges the samples that do not exist in the data set used in the model training, it is very likely to predict the wrong results. For example, there is no occupational contraindication of tetraethyl lead in the data set. The model trained with this data set cannot distinguish the occupational contraindication of tetraethyl lead as the hazard factor. In practical application, the prediction of the physical examinee with this occupational contraindication may lead to errors. The present invention simulates the decision-making process of experts by integrating the expert system, so that the model can accurately predict the samples that have not appeared in the data set. The introduction of the expert system enhances the adaptability and generalization ability of the model to unknown situations, and ensures the effectiveness and reliability of the model in practical application. In addition, the expert system also provides the interpretability of the model prediction, making the decision-making process of the model more transparent, which is convenient for medical personnel and physical examiners to understand and trust.

[0168] The present invention adopts the above technical solution and proposes an occupational health examination result diagnosis conclusion analysis and prediction model that integrates machine learning model and expert system, which is expected to improve the intelligent level of occupational health management and provide strong support for the prevention and control of occupational diseases. Through this innovative model, occupational health risks can be identified more effectively, providing a scientific basis for the formulation of preventive measures and health policies.

[0169] Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians of the art without making creative work are within the scope of protection of the present application.

Claims

1. A method for predicting occupational health examination results based on expert system and machine learning, characterized in that: The following steps are involved: Step 1, data processing: obtain the physical examination file information, clean and standardize it, and then add the number of visits information to form the original data set; some feature data of the original data set are extracted through deep learning of the original data set by the neural network model to obtain important features; The important features and the remaining feature data of the original data set together form a new feature data set; Step 2: Identify all types of hazard factors for each examinee in the new data set; perform logical judgment on each hazard factor according to the provisions of the specified occupational health standards to obtain the corresponding prediction label, and summarize it to form a decision set; mark the occupational contraindications and suspected occupational diseases in the decision set, and add the features SOD and OC corresponding to the marked results to the new data set; Step 3, hybrid model training: randomly divide the new data set that has passed the expert system link into a training set and a test set, and select two or more machine learning models to use the training set for training to obtain their respective prediction results; use the weighted voting method to weight the prediction results of the two or more models to obtain the final output result.

2. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1-1, traverse the physical examination file information and delete the redundant feature information that is not related to the physical examination conclusion; the redundant feature information is the personal information feature, and the personal information feature includes name, physical examination number, employer name, and employer region; Step 1-2, extracting hazard factor information from the physical examination file information; converting each single hazard factor in the hazard factor information into a corresponding dummy variable to form a dummy variable set, and mapping the single hazard factor name of each dummy variable to the hazard factor name classified in the specified occupational health standard; wherein the hazard factor information includes physical examination hazard factors and contact hazard factors; Step 1-3, identifying and obtaining time feature information from the physical examination file information, uniformly converting the number of months and saving it in integer form; Step 1-4, supplementing the number of visits information on the basis of the extracted information from the physical examination file information to form the original data set together; Steps 1-5, construct a neural network model, input part of the feature data of the original data set into the neural network model to perform deep learning on the original data set to extract important features; the important features and the remaining feature data of the original data set together form a new feature data set.

3. The method for predicting occupational health examination results based on expert system and machine learning according to claim 2, characterized in that: The time characteristic information in steps 1-3 includes total length of service, length of service with injuries, and start and end dates.

4. The method for predicting occupational health examination results based on expert system and machine learning according to claim 2, characterized in that: In steps 1-5, the neural network model includes an input layer, a hidden layer, and an output layer; the input layer is used to receive data with a dimension equal to the number of features in the training set; the hidden layer includes two linear layers, the first linear layer maps the input features to a 256-dimensional feature vector, and Dropout regularization is added after the first linear layer, and the dropout rate of Dropout regularization is 0.2; the second linear layer compresses the 256-dimensional feature vector outputted by the first linear layer to obtain a 128-dimensional feature vector, and the data features outputted by the second linear layer are used as important features; Each linear layer is followed by a ReLU activation function to introduce nonlinearity and enhance the expressiveness of the model; the linear layer of the output layer maps the 128-dimensional feature vector to the number of target categories.

5. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: In steps 1-5, the neural network model uses the cross entropy loss function and the Adam optimizer to train the network.

6. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: The predicted labels in step 2 include "re-examination", "suspected occupational disease", "occupational contraindications" and other diseases or abnormalities.

7. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: The specific steps of step 2 are as follows: Step 2-1, identify all types of hazard factors of the physical examination personnel in the data set: F = {f1,f2,...,f n }, n represents the number of single hazard factor characteristics, f n represents the nth single hazard factor characteristic; Traverse the single hazard factor feature set F for each physical examination record, and obtain the single hazard factor feature set G corresponding to each physical examination record. The feature set G of the i-th physical examination record is G i ; G = {g1, g2, …, g k }, k is the number of single hazard factor characteristics with a value equal to 1; Step 2-2, identify and obtain the physical examination feature t corresponding to the hazard factor f of each physical examination record, where T is the set of physical examination features t; the physical examination feature set T of the i-th physical examination record is T i :T={t1,t2,…,t d ,d is the number of physical examination features}; Step 2-3, construct a logic function H for each hazard factor f. The specific expression is as follows: Step 2-4: Establish a corresponding decision set D for a single hazard factor feature set G, D = {d1, d2, ..., d k ,d i =H(f i ,T i )},d i Denotes the single hazard factor g corresponding to the single hazard factor feature set G i The value judged under the logic function H; the physical examination feature set D of the jth physical examination record is D j : Step 2-5, mark the occupational contraindications and suspected occupational diseases in the decision set to obtain new features SOD and OC, respectively, and add the new features SOD and OC to the new data set; Among them, the feature SOD represents the expert system simulating the expert decision-making on suspected occupational diseases. 1 indicates the existence of suspected occupational diseases, and 0 indicates the absence of such diseases. The feature OC represents the judgment of occupational contraindications by the expert system simulating expert decision-making. 1 means the occupational contraindication exists, and 0 means the contraindication does not exist.

8. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: In step 3, the accuracy of the test set in each learning model is used as the evaluation indicator, and grid search is used to find the hyperparameters of each learning model.

9. The method for predicting occupational health examination results based on expert system and machine learning according to claim 1, characterized in that: The specific steps of step 3 are as follows: Step 3-1, randomly divide the new data set after the expert system into a training set and a test set in a ratio of 7:3; use the training set to train two or more machine learning models respectively, and use grid search to set the hyperparameter range of each learning model and use the accuracy of the test set predicted on the model as the evaluation index to find the hyperparameter of the model; In step 3-2, the prediction results of two or more models are weighted averaged to obtain the final output result through the weighted voting method, in which the accuracy of each model in the test set is used as the evaluation index to assign the weight to each model.

10. The method for predicting occupational health examination results based on expert system and machine learning according to any one of claims 1, 8 or 9, characterized in that: The formula for weighted voting is as follows: in, is the prediction result of the i-th model, ω i is the weight of the i-th model, and the weight is assigned to each model according to its performance on the test set.

Images