Heart disease occupational health risk assessment system based on RF-MLP model

Abstract

The invention relates to a heart disease occupational health risk assessment system based on an RF-MLP model, and belongs to the technical field of health risk prediction. In order to solve the problem that prediction performance is limited due to the fact that an existing method is insufficient in the aspects of feature selection and model optimization, the invention provides a collaborative prediction scheme fusing recursive feature elimination (RFE), a random forest (RF) and a multi-layer perceptron (MLP). According to the technical scheme, the method comprises the steps that a data collecting and preprocessing module cleans and normalizes physiological indexes; the feature selection module calculates feature importance scores through an RFE-RF algorithm to screen key indexes; the random forest optimization module constructs a decision tree cluster; the MLP module is used for realizing nonlinear mapping through ReLU and Sigmoid functions; and the model fusion module integrates the output of the double models to generate final prediction. The system remarkably improves the prediction precision and generalization ability, provides high-robustness support for early warning of heart diseases of occupational people, and promotes disease prevention to be converted from passive response to active intervention.

Description

Technical Field

[0001] This invention belongs to the field of health risk prediction technology, and relates to a cardiac occupational health risk assessment system based on a Random Forest-Multi-Layer Perceptron (RF-MLP) fusion model. Background Technology

[0002] In recent years, health risk prediction technologies based on machine learning and deep learning have made significant progress, effectively predicting disease risk through the analysis of medical data. However, existing methods still have shortcomings in feature selection and model optimization:

[0003] (1) Limitations of feature selection: Traditional algorithms have difficulty in comprehensively evaluating the contribution of features to disease prediction, resulting in redundant features interfering with model accuracy;

[0004] (2) Weak model generalization ability: Single models (such as logistic regression or support vector machine) are difficult to handle the high-dimensional nonlinear relationships of clinical data and are prone to overfitting;

[0005] (3) Insufficient prediction robustness: It lacks the ability to adaptively model complex physiological indicators (such as dynamic blood pressure and blood glucose fluctuations) in occupational health scenarios.

[0006] The aforementioned shortcomings make it difficult for existing technologies to meet the requirements of high accuracy and strong generalization ability for early warning of heart disease. Especially in the field of occupational health assessment, it is necessary to take into account both individual differences and group risk characteristics, and there is an urgent need for a new predictive model that integrates multi-dimensional analysis. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a cardiac occupational health risk assessment system based on the RF-MLP model, comprising six main steps: data acquisition and preprocessing, feature selection and dimensionality reduction, random forest algorithm optimization, multilayer perceptron model construction and training, model fusion prediction, and model evaluation. By fully utilizing clinical data in clinical databases, this method can extract key health indicators, optimize model performance, and provide scientific support for the early prediction and prevention of cardiac disease.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A cardiac occupational health risk assessment system based on the RF-MLP model includes a data acquisition and preprocessing module, a feature selection and dimensionality reduction module, a random forest optimization module, a multi-layer perceptron (MLP) model construction and training module, a model fusion prediction module, and a model evaluation module.

[0010] The output of the data acquisition and preprocessing module is connected to the input of the feature selection and dimensionality reduction module, the output of the feature selection and dimensionality reduction module is connected to the input of the random forest optimization module, the output of the random forest optimization module is connected to the input of the MLP model building and training module, the output of the MLP model building and training module is connected to the input of the model fusion prediction module, and the output of the model fusion prediction module is connected to the input of the model evaluation module.

[0011] The system is used to predict heart disease risk using clinical data.

[0012] Furthermore, the data acquisition and preprocessing module includes:

[0013] The data acquisition unit is used to collect physiological index data, including blood pressure, blood sugar, heart rate, and body mass index (BMI).

[0014] The data cleaning unit is used to remove outliers, including records with a heart rate greater than 220 bpm or a blood glucose level lower than 2 mmol / L.

[0015] Missing value imputation cells are used to fill continuous data with the mean or median. The imputation formula is:

[0016] x fill =mean(x) or x fill =median(x)

[0017] Where, x fill Here are the filled values, mean(x) is the mean of feature x, and median(x) is the median of feature x;

[0018] Normalization unit, used to apply the Min-Max normalization formula:

[0019]

[0020] Among them, X norm The value is the normalized value, and X is the original data value. min For the minimum value, X max It is the maximum value;

[0021] The category label transformation unit is used to convert the target variable into a binary category label y∈{0,1}, where 0 represents no risk and 1 represents risk.

[0022] Furthermore, the feature selection and dimensionality reduction module uses Recursive Feature Elimination (RFE) combined with the Random Forest (RF) algorithm (RFE-RF algorithm), including:

[0023] The recursive feature elimination unit is used to recursively delete the feature with the smallest contribution.

[0024] Feature importance score calculation unit, used to calculate feature importance scores:

[0025]

[0026] Where FI(f) is the importance score of feature f, T is the total number of decision trees in the random forest, and ΔI t (f) represents the decrease in the Gini index after feature f splits in the t-th tree.

[0027] Furthermore, the feature selection and dimensionality reduction module also includes a feature subset evaluation unit, which is used to evaluate the performance of the feature subset by calculating the area under the curve (AUC) and output the best feature subset and its number N to the random forest optimization module.

[0028] Furthermore, the random forest optimization module receives the optimal feature subset and its number N output by the feature selection and dimensionality reduction module, and constructs T decision trees based on this subset, including:

[0029] Bootstrap sampling units are used for random sampling; the sample selection formula is:

[0030] Bootstrap Sample={x1,x2,...,x n}x i ~Training Set

[0031] Where, x i The sample is the Training Set;

[0032] The decision tree construction unit uses either the Gini index or information gain as the splitting criterion, where the formula for the Gini index is:

[0033]

[0034] Where G is the Gini index, p i This represents the proportion of samples of class i in the current node, where k is the number of classes.

[0035] Furthermore, the information gain formula is:

[0036]

[0037] Where IG is the information gain, H(D) is the entropy of the current dataset D, and D i For the subset after splitting, |D i| represents the number of samples in the subset, and |D| represents the total number of samples.

[0038] Furthermore, the MLP model construction and training module includes:

[0039] The input layer is used to receive N important features;

[0040] Hidden layers, used to activate the Rectified Linear Unit (ReLU) function:

[0041] f(x) = max(0,x)

[0042] Where f(x) is the output of the activation function;

[0043] The output layer is used to output the predicted probabilities using the Sigmoid activation function.

[0044]

[0045] Where σ(x) is the prediction probability;

[0046] Loss function unit, used to employ the cross-entropy loss function:

[0047]

[0048] Where L is the loss value, y i For actual labels, Here, m represents the predicted probability of the model, and m is the number of samples.

[0049] Furthermore, the MLP model construction and training module also includes:

[0050] Adam optimizer unit, used for parameter updates:

[0051]

[0052] Where, θ t+1 For the updated model parameters, θ t Here are the model parameters for the current iteration step, where α is the learning rate and m is the number of steps. t v is the exponential moving average of the gradient. t ε is the exponential moving average of the squared gradient, and ε is the numerical stability constant.

[0053] The learning rate decay unit is used to dynamically adjust the learning rate.

[0054]

[0055] Where, α t Let α be the learning rate for the t-th iteration, α0 be the initial learning rate, λ be the decay rate, and t be the number of iterations.

[0056] Furthermore, the model fusion prediction module is used to fuse random forest and MLP models, and the final prediction formula is:

[0057]

[0058] in, For the prediction result, X is the feature subset output by the random forest optimization module, W1 and W2 are weight matrices, and b1 and b2 are bias parameters.

[0059] Furthermore, the model evaluation module is used to calculate evaluation indicators, including:

[0060] Accuracy:

[0061]

[0062] Recall:

[0063]

[0064] F1 score:

[0065]

[0066] Area Under the Curve (AUC);

[0067] Among them, TP represents true positive, TN represents true negative, FP represents false positive, FN represents false negative, and Precision represents accuracy.

[0068] The beneficial effects of this invention are as follows:

[0069] This invention significantly improves the performance and practicality of heart disease prediction in occupational health assessment through an innovative RF-MLP model fusion architecture and systematic processing flow. The beneficial effects are reflected in multiple dimensions, including fundamental optimization of prediction accuracy, substantial enhancement of model generalization ability, systematic improvement of data processing efficiency, and comprehensive expansion of practical application value. The specific effects are described below, starting with the core advantages.

[0070] (1) This invention employs recursive feature elimination combined with a random forest algorithm to dynamically screen key clinical indicators (such as blood pressure, blood glucose, and heart rate), eliminate redundant interference, and ensure that the model focuses on high-contribution features. This mechanism quantifies feature value through feature importance assessment, avoiding the blindness of feature selection in traditional methods, thereby achieving more reliable prediction results in complex physiological data scenarios. The model fusion architecture (random forest and multilayer perceptron collaboration) effectively captures the high-dimensional nonlinear relationships of the data, reduces the risk of overfitting, and improves adaptability to dynamic occupational health indicators.

[0071] (2) The multilayer perceptron module utilizes ReLU and Sigmoid activation functions to achieve nonlinear mapping, combined with the Bootstrap sampling mechanism of random forest, to construct a highly robust prediction framework. This dual-model complementary mechanism optimizes convergence efficiency during training through dynamic learning rate adjustment, ensuring that the model maintains stable performance across different occupational groups (such as high-risk occupational groups). The systematic training process avoids the limitations of a single model, making the prediction results more universal and consistent.

[0072] (3) From data acquisition to normalization processing, this invention integrates automated cleaning and filling units to efficiently handle missing and outlier values ​​(such as abnormal heart rate or blood glucose records), ensuring the quality of input data. The recursive feature elimination algorithm simplifies feature dimensions and reduces computational complexity, while maintaining data scale uniformity through Min-Max normalization, accelerating model training. This optimization enables the system to respond faster and consume fewer resources in real-time occupational health monitoring.

[0073] (4) The system is designed for occupational health scenarios, providing early warning and prevention support, and enabling flexible deployment through a B / S architecture (such as a custom database for medical institutions). A multi-dimensional assessment module (covering accuracy, recall, F1 score, and AUC) ensures the credibility of the prediction results and provides a scientific basis for clinical intervention. Role-based management of the user platform and the administrator platform enhances the system's usability and security, supporting large-scale occupational health management needs.

[0074] (5) RF-MLP model is an abbreviation for "Random Forest-Multilayer Perceptron Fusion Model". This model integrates the random forest algorithm with multilayer perceptron technology, achieving high-precision disease risk prediction through a complementary mechanism of feature selection and nonlinear mapping. Among them:

[0075] Random forest: Used for key feature selection and preliminary decision-making, improving the stability and interpretability of the model.

[0076] Multilayer perceptron: Responsible for deep nonlinear modeling, capturing complex data relationships through multilayer neural networks.

[0077] The fused model can efficiently process physiological indicators such as blood pressure and heart rate in occupational health scenarios and output reliable probabilities of heart disease risk.

[0078] (5) The RF-MLP fusion process includes:

[0079] Feature selection stage: The random forest algorithm evaluates feature importance through recursive feature elimination and filters key indicators (such as systolic blood pressure and Thallium value).

[0080] Deep prediction stage: The multilayer perceptron receives the filtered features and uses ReLU and Sigmoid activation functions to perform nonlinear transformations to generate the final risk probability.

[0081] In the prediction system of this invention, RF-MLP serves as the core engine, driving the entire process from data preprocessing to risk assessment.

[0082] (6) The RF-MLP model significantly improves prediction accuracy and generalization ability through the complementarity of two models:

[0083] Enhanced robustness: Random forests handle high-dimensional feature noise, and multilayer perceptrons optimize the fitting of complex patterns, reducing the risk of overfitting.

[0084] Support occupational health applications: Provide healthcare professionals with visual risk assessment tools to enable early warning and personalized intervention.

[0085] This model is the core innovation of this invention, which breaks through the limitations of traditional single algorithms and promotes the shift of heart disease prediction from passive response to proactive prevention.

[0086] In summary, this invention, through feature selection optimization, model fusion innovation, and systematic process design, improves prediction reliability while providing an efficient and scalable solution for occupational heart disease risk assessment, ultimately promoting disease prevention from passive response to proactive intervention.

[0087] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. Attached Figure Description

[0088] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0089] Figure 1 The flowchart of the RF-MLP model prediction method shows the overall process of the model and the relationship between each step.

[0090] Figure 2 The diagram illustrates the principle of feature selection in the random forest algorithm, showcasing the importance evaluation method during the feature selection process.

[0091] Figure 3 A ranking chart of clinical features shows the contribution of different clinical features to disease risk prediction.

[0092] Figure 4 The diagram illustrates the working principle of the Multilayer Perceptron (MLP) algorithm, showcasing its hierarchical structure and operating principle.

[0093] Figure 5 The graph shows the relationship between the accuracy, loss value, and number of iterations on the training and validation sets of the model, illustrating the changes in accuracy and loss value during the training process.

[0094] Figure 6 The ROC curve performance graphs for each model are shown, illustrating the performance of different models under the ROC curve and comparing their classification abilities.

[0095] Figure 7 A diagram of the user platform's functional modules;

[0096] Figure 8 Diagram of the administrator platform's functional modules;

[0097] Figure 9 System architecture diagram;

[0098] Figure 10 This is a schematic diagram of the administrator's side functions;

[0099] Figure 11 This is a schematic diagram of the functions for individual users.

[0100] Figure 12 A diagram illustrating the use of satisfaction metrics in the model;

[0101] Figure 13 This diagram illustrates enabling the training model for the model.

[0102] Figure 14 This is a diagram showing that the model has been successfully enabled. Detailed Implementation

[0103] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0104] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0105] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0106] The specific implementation of this invention will be described through the following steps: data acquisition and preprocessing, feature selection and dimensionality reduction, random forest algorithm optimization, multilayer perceptron (MLP) model construction and training, model fusion prediction, and model evaluation and validation. The implementation steps are detailed below:

[0107] S1: Data Acquisition and Preprocessing

[0108] Data Acquisition: Data is collected from the UCI machine learning platform, which has been used by numerous researchers in scientific studies. Collected physiological indicators include, but are not limited to, blood pressure (systolic / diastolic), blood glucose, heart rate, and body mass index (BMI). Data sources can include external health monitoring platforms, as shown in Table 1.

[0109] Table 1

[0110]

[0111]

[0112] Data cleaning: Data cleaning is a crucial step in ensuring data quality. First, outliers that do not meet the criteria are removed. For example, records with a heart rate exceeding 220 bpm or blood glucose levels below 2 mmol / L need to be deleted to ensure data accuracy.

[0113] Missing value imputation: For missing data, the mean or median is used for imputation. Specifically, for continuous data (such as blood pressure and heart rate), the mean or median of the feature can be used for imputation.

[0114] Normalization: To ensure the stability of the model during training, standardization or normalization methods are used to process numerical data. The following formula is used for data normalization:

[0115]

[0116] Among them, X norm The value is the normalized value, and X is the original data value. min For the minimum value, X max It is the maximum value;

[0117] Category label conversion: Convert the target variable (whether a cardiac event occurred) into a binary category label, where 0 represents no risk and 1 represents risk.

[0118] S2: Feature Selection and Dimensionality Reduction

[0119] Recursive Feature Elimination and Random Forest Algorithm (RFE-RF): This algorithm combines Recursive Feature Elimination (RFE) and Random Forest (RF) for feature selection. First, a Random Forest model is used to evaluate the importance of each feature, and then the RFE method recursively removes features with the lowest contribution. For each feature subset, the model performance is evaluated by calculating the AUC value, and the optimal feature subset is finally selected. The optimal feature subset and its size N will serve as the input feature dimension for the Random Forest optimization module.

[0120] S3: Random Forest Algorithm Optimization

[0121] Decision tree construction: Multiple decision trees are constructed using the random forest algorithm. The sample selection method for each tree is as follows:

[0122] Random sampling with replacement is performed from the training set.

[0123] The implementation process of this sampling method in random forest is as follows:

[0124] (1) Data preparation

[0125] The original training set contains N records;

[0126] Set the sample size n for each decision tree (n≤N).

[0127] (2) Sampling with replacement

[0128] Execute the loop n times:

[0129] A sample x is randomly selected from the training set. i ;

[0130] Record x i Then it was put back into the original dataset.

[0131] Generate a Bootstrap sample set:

[0132] Bootstrap Sample={x1,x2,...,x n}x i ~Training Set

[0133] (3) Decision tree construction

[0134] Train a single decision tree independently using the sampling results.

[0135] Repeat T times to build the forest.

[0136] S4: Construction and Training of Multilayer Perceptron (MLP) Model

[0137] Model Structure: The multilayer perceptron model consists of an input layer, hidden layers, and an output layer. The input layer receives key features selected by a random forest, the hidden layers use activation functions for non-linear mapping, and the output layer uses activation functions to output predicted probabilities.

[0138] Activation and Loss Functions: The hidden layer uses the ReLU (Rectified Linear Unit) activation function, and the output layer uses the Sigmoid activation function. The loss function is the cross-entropy loss function, as shown in the following formula:

[0139]

[0140] Where L is the loss value, y i For actual labels, Here, m represents the predicted probability of the model, and m is the number of samples.

[0141] S5: Model Training and Optimization

[0142] Optimizer selection: The Adam optimizer is used for parameter updates, as shown in the following formula:

[0143]

[0144] Where, θ t+1 For the updated model parameters, θt Here are the model parameters for the current iteration step, where α is the learning rate and m is the number of steps. t v is the exponential moving average of the gradient. t ε is the exponential moving average of the squared gradient, and ε is the numerical stability constant.

[0145] Learning rate decay: To improve training efficiency, a dynamic learning rate adjustment method is used, gradually reducing the learning rate. The learning rate decay formula is as follows:

[0146] η t+1 =η0·decay rate t

[0147] Where η0 is the initial learning rate, decay rate is the decay rate, and t is the current iteration number.

[0148] S6: Model Fusion and Prediction

[0149] Model fusion: The final prediction is achieved by fusing the outputs of Random Forest and Multilayer Perceptron. Features selected by Random Forest are fed as input data to the Multilayer Perceptron model, which outputs the final prediction result.

[0150]

[0151] in, This is the predicted result.

[0152] Figure 1 The flowchart of the RF-MLP model prediction method shows the overall process of the model and the relationship between each step.

[0153] Figure 2 The diagram illustrates the principle of feature selection in the random forest algorithm, showcasing the importance evaluation method during the feature selection process.

[0154] Figure 3 A ranking chart of clinical features shows the contribution of different clinical features to disease risk prediction.

[0155] Figure 4 The diagram illustrates the working principle of the Multilayer Perceptron (MLP) algorithm, showcasing its hierarchical structure and operating principle.

[0156] Figure 5 The graph shows the relationship between the accuracy, loss value, and number of iterations on the training and validation sets of the model, illustrating the changes in accuracy and loss value during the training process.

[0157] Figure 6 The ROC curve performance graphs for each model are shown, illustrating the performance of different models under the ROC curve and comparing their classification abilities.

[0158] This system can be divided into two main parts: a "user platform" and a "management platform." The user platform is designed for general users, providing patients with convenient and easy-to-understand predictive services and accurate predictions. The management platform is primarily for system administrators, providing users with a data management tool that integrates data collection, storage, cleaning, and analysis, and includes functions such as model selection, parameter adjustment, and performance evaluation. To ensure data security and system stability, the management platform must also have appropriate access control functionality. The following are details regarding the functional requirements of these two subsystems.

[0159] User platform functions such as Figure 7 As shown. The administrator platform functions are as follows: Figure 8 As shown.

[0160] 1. System Architecture

[0161] This invention's prediction system adopts a B / S (Browser / Server) architecture. Users can access the system's front-end interface through a browser and securely log in using their personal accounts. The B / S architecture provides the system with great flexibility, allowing medical institutions to easily choose between using the standard database provided by the system developers or a self-created database, based on their specific needs. These adjustments can be made simply by modifying relevant configuration files. The system is deployed on a dedicated server, ensuring service stability and security. Furthermore, users can conveniently access the cardiac prediction system via a webpage from anywhere with internet access. This design not only provides efficient prediction services but also greatly enhances user convenience. The system architecture diagram of this invention is shown below. Figure 9 As shown.

[0162] 2. System Roles

[0163] 2.1 Administrator User Functions

[0164] For administrator users, the functional design primarily focuses on comprehensive management of various types of data within the system, including but not limited to personal information, data records, medical records, and model training data. Administrators possess the highest privileges, ensuring the integrity, security, and accuracy of the data. Figure 10 As shown.

[0165] 2.2 Functions for Individual Resident Users

[0166] Individual residents primarily focus on viewing and updating their personal health information, as well as interacting with doctors. The system design fully considers user convenience and privacy, ensuring that individual residents can easily manage their health data, communicate effectively with doctors, and input their physical examination data into the system based on suggested indicators for heart disease prediction and view the results. The user interface includes features such as... Figure 11As shown.

[0167] To meet the specific needs of different roles during operation, the core principle of this system design is to manage user roles meticulously and separately. Through this design, the invention ensures that each role can only access functions and data relevant to their responsibilities, thereby guaranteeing system security while significantly improving operational efficiency.

[0168] In this system, the invention primarily defines two user roles: administrators and individual resident users. Administrators, as senior users, are granted comprehensive management privileges. They can monitor the entire system and manage user accounts, including creating, modifying, and deleting users. Simultaneously, system administrators can also fully manage and maintain various data and models to ensure data accuracy and model execution efficiency.

[0169] Individual residents are the system's regular users, primarily interacting through the system's interface by inputting their relevant information to obtain predictions of their heart disease risk. Based on the user's input data and the model's predictions, the system provides personalized health advice and reminders. Individual residents can also view their historical records and health trends for better health management. Detailed role classifications and precise access controls ensure high system efficiency and accurate data management. Whether for administrators or individual residents, everyone can find suitable entry points and functional modules within the system to easily complete various tasks or health management duties.

[0170] Administrators can easily access the system's built-in predictive models through this module to quickly analyze and predict medical data. This feature not only improves the efficiency and accuracy of medical decision-making but also provides strong data support for researchers, driving advancements in medical technology.

[0171] If the model's prediction accuracy fails to reach the preset 90% key performance indicator after training, the administrator will not be satisfied with the current performance. The administrator will carefully review and adjust relevant input parameters, such as the dataset, number of iterations, and number of hidden layers, to improve the model's predictive capabilities. This process may involve multiple trials and adjustments until the model's prediction accuracy significantly improves and meets the administrator's satisfaction standard. The administrator will then quickly deploy this refined new model to the user prediction platform so that users can experience more accurate and reliable prediction services. Figure 12 As shown. Enabling model display is as follows. Figure 13 As shown, successful model activation is indicated as follows. Figure 14 As shown.

[0172] Example 1: Data Preprocessing and Feature Selection Process

[0173] 1.1 Data Acquisition: 2000 clinical records were obtained from the UCI machine learning platform, including fields such as age, gender, systolic / diastolic blood pressure, blood glucose, BMI, cholesterol, and Thallium value.

[0174] 1.2 Outlier Cleaning:

[0175] Automatically delete records with a heart rate > 220 bpm (e.g., ID 1002 heart rate 234 bpm).

[0176] Exclude abnormal data with blood glucose levels <2 mmol / L (e.g., ID 1001 blood glucose 1.8 mmol / L).

[0177] 1.3 Handling Missing Values:

[0178] Use the mean-fill formula to fill in missing values ​​in the blood glucose field (e.g., ID 1003):

[0179] x fill =mean(FBS)=0.21

[0180] 1.4 Normalization and Label Transformation:

[0181] The blood pressure field was processed using Min-Max normalization:

[0182]

[0183] Convert the "Heart Disease" column to a binary label ("Presence" → 1, "Absence" → 0).

[0184] 1.5 Recursive Feature Elimination (RFE-RF):

[0185] Random forests calculate feature importance scores:

[0186]

[0187] Sort by score (Thallium > Systolic Blood Pressure > Age > BMI)

[0188] The optimal feature subset (N=5) was determined by AUC evaluation.

[0189] Example 2: RF-MLP Model Training and Optimization Process

[0190] 2.1 Random Forest Construction:

[0191] Generate T = 100 decision trees

[0192] Bootstrap sampled training set (80% of samples):

[0193] Sample = {x1, x3, x5, ..., x} 1998}

[0194] Use the Gini index to divide nodes (k=2 classes):

[0195] G = 1 - (0.7) 2 +0.3 2 ) = 0.42

[0196] 2.2 MLP Model Structure:

[0197] Input layer: Receives the 5-dimensional feature vector output by RFE-RF.

[0198] Hidden layers (2 layers):

[0199] Number of neurons: 128 → 64

[0200] ReLU activation function: f(x) = max(0, 0.86) = 0.86

[0201] Output layer: Sigmoid function output probability:

[0202]

[0203] 2.3 Dynamic Training Optimization:

[0204] The initial learning rate α0 = 0.001, and the decay rate λ = 0.01.

[0205] The learning rate is updated in the 100th iteration.

[0206]

[0207] The Adam optimizer updates the weight parameter W1.

[0208] Example 3: Model Fusion Prediction and Occupational Health Assessment Process

[0209] 3.1 Dual-model fusion prediction:

[0210] Random forest outputs feature subset input to MLP

[0211] The fusion prediction formula calculates the individual risk value:

[0212]

[0213] Threshold setting:

[0214] 3.2 Multi-dimensional performance verification:

[0215] Calculate the metrics for the test set (500 records):

[0216] Accuracy: Recall rate: AUC = 0.93

[0217] 3.3 Occupational health application scenarios:

[0218] Input the truck driver's medical examination data (age 45, systolic blood pressure 142 mmHg, Thallium = 6), and the system will output the risk probability. → Triggers a "high-risk occupational exposure" warning;

[0219] Generate health recommendations: blood pressure monitoring + ECG follow-up.

[0220] Example 4: System Deployment and Role Management Process

[0221] 4.1 System Architecture Deployment:

[0222] Deployed on cloud servers using a B / S architecture

[0223] Healthcare institutions can switch databases (MySQL → PostgreSQL) via configuration files.

[0224] 4.2 Administrator Platform Operations:

[0225] Model Management: Upload new training set and update RF-MLP parameters

[0226] Access control: Setting up doctor account access to patient risk reports

[0227] Data monitoring: Real-time detection of abnormal input (e.g., heart rate > 200 bpm)

[0228] 4.3 User platform interaction:

[0229] Individual users input their health checkup indicators;

[0230] Dynamically displays risk level (low / medium / high);

[0231] Generate historical trend charts (e.g., blood pressure changes over three consecutive months).

[0232] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cardiac occupational health risk assessment system based on the RF-MLP model, characterized in that: It includes modules for data acquisition and preprocessing, feature selection and dimensionality reduction, random forest optimization, multilayer perceptron (MLP) model construction and training, model fusion and prediction, and model evaluation. The output of the data acquisition and preprocessing module is connected to the input of the feature selection and dimensionality reduction module, the output of the feature selection and dimensionality reduction module is connected to the input of the random forest optimization module, the output of the random forest optimization module is connected to the input of the MLP model building and training module, the output of the MLP model building and training module is connected to the input of the model fusion prediction module, and the output of the model fusion prediction module is connected to the input of the model evaluation module. The system is used to predict heart disease risk using clinical data.

2. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The data acquisition and preprocessing module includes: The data acquisition unit is used to collect physiological index data, including blood pressure, blood sugar, heart rate and body mass index (BMI); The data cleaning unit is used to remove outliers, including records with a heart rate greater than 220 bpm or a blood glucose level lower than 2 mmol / L. Missing value imputation cells are used to fill continuous data with the mean or median. The imputation formula is: x fill =mean(x) or x fill =median(x) Where, x fill Here are the filled values, mean(x) is the mean of feature x, and median(x) is the median of feature x; Normalization unit, used to apply the Min-Max normalization formula: Among them, X norm The value is the normalized value, and X is the original data value. min For the minimum value, X max It is the maximum value; The category label transformation unit is used to convert the target variable into a binary category label y∈{0,1}, where 0 represents no risk and 1 represents risk.

3. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The feature selection and dimensionality reduction module uses Recursive Feature Emission (RFE) combined with the Random Forest (RF) algorithm, including: The recursive feature elimination unit is used to recursively delete the feature with the smallest contribution. Feature importance score calculation unit, used to calculate feature importance scores: Where FI(f) is the importance score of feature f, T is the total number of decision trees in the random forest, and ΔI t (f) represents the decrease in the Gini index after feature f splits in the t-th tree.

4. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 3, characterized in that: The feature selection and dimensionality reduction module also includes a feature subset evaluation unit, which evaluates the performance of the feature subset by calculating the area under the curve (AUC) and outputs the best feature subset and its number N to the random forest optimization module.

5. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The random forest optimization module receives the optimal feature subset and its number N output by the feature selection and dimensionality reduction module, and constructs T decision trees based on this subset, including: Bootstrap sampling units are used for random sampling; the sample selection formula is: Bootstrap Sample={x1,x2,...,x n }x i ~Training Set Where, x i The sample is the Training Set; The decision tree construction unit uses either the Gini index or information gain as the splitting criterion, where the formula for the Gini index is: Where G is the Gini index, p i This represents the proportion of samples of class i in the current node, where k is the number of classes.

6. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 5, characterized in that: The information gain formula is: Where IG is the information gain, H(D) is the entropy of the current dataset D, and D i For the subset after splitting, |D i | represents the number of samples in the subset, and |D| represents the total number of samples.

7. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The MLP model construction and training module includes: The input layer is used to receive N important features; Hidden layers used with the modified linear unit ReLU activation function: f(x) = max(0,x) Where f(x) is the output of the activation function; The output layer is used to output the predicted probabilities using the Sigmoid activation function. Where σ(x) is the prediction probability; Loss function unit, used to employ the cross-entropy loss function: Where L is the loss value, y i For actual labels, Here, m represents the predicted probability of the model, and m is the number of samples.

8. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 7, characterized in that: The MLP model construction and training module also includes: Adam optimizer unit, used for parameter updates: Where, θ t+1 For the updated model parameters, θ t Here are the model parameters for the current iteration step, where α is the learning rate and m is the number of steps. t v is the exponential moving average of the gradient. t ε is the exponential moving average of the squared gradient, and ε is the numerical stability constant. The learning rate decay unit is used to dynamically adjust the learning rate. Where, α t Let α be the learning rate for the t-th iteration, α0 be the initial learning rate, λ be the decay rate, and t be the number of iterations.

9. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The model fusion prediction module is used to fuse random forest and MLP models, and the final prediction formula is: in, For the prediction result, X is the feature subset output by the random forest optimization module, W1 and W2 are weight matrices, and b1 and b2 are bias parameters.

10. The occupational health risk assessment system for heart disease based on the RF-MLP model according to claim 1, characterized in that: The model evaluation module is used to calculate evaluation indicators, including: Accuracy: Recall: F1 score: The area under the curve (AUC) Among them, TP represents true positive, TN represents true negative, FP represents false positive, FN represents false negative, and Precision represents accuracy.

Images