Global suicide rate prediction method based on multi-dimensional parameters and random forest model

Abstract

The invention discloses a global suicide rate prediction method based on multi-dimensional parameters and a random forest model, and belongs to the field of public health prediction. The method comprises the following specific steps: step 1, determining a national scope and a time sequence according to statistical data of websites such as a world health organization and a world bank; step 2, acquiring multi-dimensional data; 3, constructing a machine learning model and adjusting and optimizing the model; 4, predicting the suicide rate of the future global country scale; and step 5, result visualization. According to the method, multi-dimensional parameters such as natural environment, social economy, population structure and disease factors are integrated, a high-precision prediction model is constructed by using a machine learning algorithm, and a scientific basis can be provided for a global suicide prevention strategy.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of public health and epidemic prediction, and specifically relates to a global suicide rate prediction method based on multi-dimensional parameters and a random forest model. BACKGROUND

[0002] Suicide is a major global public health problem, causing approximately 800,000 deaths annually. Accurate prediction of suicide rate trends is crucial for developing effective prevention strategies. However, suicide behavior is influenced by a variety of complex factors, including natural environment, socioeconomic, population structure, and mental health, among other multi-dimensional factors, posing significant challenges for traditional prediction methods.

[0003] Currently, global-scale suicide rate prediction mainly relies on time series trend extrapolation and traditional regression models: 1. Time series analysis: based on historical suicide rate data for trend extrapolation, but ignores changes in influencing factors; 2. Regression model: considers limited influencing factors, making it difficult to capture complex nonlinear relationships; 3. Expert or national disease control organization statistical assessment: relies on subjective judgment and lacks objective data support.

[0004] The main defects of the prior art are: unable to integrate multi-source heterogeneous data; model generalization ability is insufficient, making it difficult to adapt to different national characteristics; lack of dynamic updating mechanism, unable to timely reflect the impact of environmental and social changes. For example, Chinese patent CN201910358712.X discloses a "suicide risk prediction method based on multiple factors", but it only considers local social and economic factors, does not integrate global-scale natural environment parameters and comprehensive disease factor data, and the model prediction accuracy and scope are limited. SUMMARY

[0005] To address the above problems in the prior art, the present application aims to provide a global suicide rate prediction method based on multi-dimensional parameters and a random forest model, which integrates multi-dimensional data modeling to achieve spatiotemporal distribution prediction of suicide rates worldwide.

[0006] The present application provides the following technical solution: a global suicide rate prediction method based on multi-dimensional parameters and a random forest model, comprising the following steps: Step 1: Determine the list of countries and time series included in the study range; Step 2: Collect annual suicide rate data for each country as the target variable; Step three, build a multi-dimensional prediction index system, obtain the modeling basis data of each country in the research scope, use the random forest regression model in the machine learning library, select the optimal model parameters through model training and verification, and realize the updateable global suicide rate prediction model through updating the modeling basis data; Step four, predict the future suicide rate of the target country; Step five, visualize the spatial distribution map of the global suicide rate prediction results and the suicide rate trend chart of each country.

[0007] Further, in step two, the suicide rate data of each country needs to be labeled with data source, statistical method and adjustment coefficient to ensure data comparability.

[0008] Further, in step three, the modeling basis data includes natural environment parameters, economic parameters, social parameters and disease factors; based on the acquisition of multi-variable modeling basis data, when building a multi-dimensional prediction index system, introduce parameter interaction items to enhance the explanation depth and prediction accuracy of the model. The interaction item captures the synergistic or antagonistic effect between variables, revealing the action mechanism reflected by multiple variables. For example, the interaction between extreme high temperature in natural environment parameters and public health investment in economic parameters can evaluate how economic resources adjust the impact of climate events on public health; the interaction item between Gini coefficient in social parameters and malnutrition rate in disease factors can analyze the amplification effect of social and economic inequality on nutritional health outcomes. In addition, the interaction item between per capita GDP in economic parameters and CO2 emissions in natural environment parameters can verify the moderating effect of economic development level on environmental health effects, providing a basis for society to develop differentiated climate policies. By systematically introducing the above interaction items, this index system not only can more accurately simulate the multi-factor dynamic coupling relationship in the real world, but also can identify key intervention nodes to provide theoretical support for optimizing resource allocation and developing targeted measures, thereby improving the scientific value and decision-making reference significance of the model.

[0009] Further, in step three, the specific implementation process of the global suicide rate prediction model is as follows: 3.1) Select a machine learning model: select a regression model suitable for suicide rate multi-factor prediction from the machine learning library; 3.2) Data preprocessing: convert parameters of different dimensions into comparable values through standardization processing; 3.3) Establish a data set: the input data set includes feature values X and target values Y, target values Y represent the suicide rate of a country in a year, and feature values X represent multi-dimensional factors affecting Y; 3.4) Splitting of the sample: The data set is randomly divided into training samples Xtrain, Ytrain and test samples Xtest, Ytest in proportion; the training samples are used to learn the model of historical data; the test samples are used to evaluate the prediction performance of the model, and the data of a country is kept complete during the division, that is, all the data of a country in all years are all used as the training set or the test set, so as to improve the generalization ability of the model while ensuring randomness; 3.5) Training model: different hyperparameter combinations are set, the training model is trained by inputting the training samples Xtrain and Ytrain under different combination conditions; meanwhile, the Xtrain of the training sample and the Xtest of the test sample are input into the model, and the predicted results Ypre_train and Ypre_test of the model are compared with Ytrain and Ytest respectively, and the coefficient of determination and the root mean square error are used as the evaluation parameters, and the model with the smallest error and the hyperparameters are selected.

[0010] Further, the step 3.2) is specifically as follows: The continuous parameters are standardized, the classification variables are one-hot encoded, and the features of skew distribution are logarithmically transformed.

[0011] Further, in the step five, the global suicide rate prediction result spatial distribution map of the same period is distinguished by color according to different suicide rate levels; and the suicide rate trend chart of the same country in different periods presents the time change trend through the line chart.

[0012] The drawing process of the global suicide rate prediction result spatial distribution map and the suicide rate trend chart is as follows: First, the data set is input into the model, and the two training set and validation set coefficient of determination and root mean square error maps are obtained Figure 2 by the operation of the model. The prediction accuracy and generalization ability of the model are determined by judging the performance of the training set and the validation set coefficient of determination, the gap between them and the root mean square error of the training set and the validation set. The high training set coefficient of determination and the small gap between the training set coefficient of determination and the validation set coefficient of determination, and the small root mean square error of the training set and the validation set prove that the prediction accuracy of the model is high, the generalization ability is strong, and the model can be used for prediction. When the prediction accuracy and generalization ability of the model meet the standard, the future parameters (the parameters not collected are defaulted to take the parameters of the latest year) are input into the model for future global suicide rate prediction, and the future results predicted by the model are input into Arcgis for drawing the global suicide rate prediction result spatial distribution map and the suicide rate trend chart.

[0013] Compared with the prior art, the beneficial effects of the present application are as follows: The application integrates four influencing factors of natural environment, social economy, population structure and disease factors, adopts a random forest algorithm, can capture complex nonlinear relationships and interactions, realizes synchronous prediction of multiple countries in the global range, and provides a scientific basis for suicide prevention. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of a global suicide rate prediction method based on multi-source data and a machine learning model in the specific embodiment of the application is shown in the figure. Figure 2 In the application, a prediction and observation sample distribution graph of a training set and a test set of the random forest model training data is shown in the figure; a is a prediction and observation sample distribution graph of the training set, the determination coefficient is 0.9466, and the root mean square error is 1.7488; b is a prediction and observation sample distribution graph of the test set, the determination coefficient is 0.9029, and the root mean square error is 1.9843. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the application more clear and understandable, the application is further described in detail below in combination with the drawings and examples in the specification. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0016] On the contrary, the application covers any substitution, modification, equivalent method and scheme defined by the claims on the essence and scope of the application. Further, in order to make the public better understand the application, some specific details are described in detail in the following detailed description of the application. The application can also be completely understood without the description of these details by those skilled in the art.

[0017] The application provides a global suicide rate prediction method based on multi-source data and a machine learning model, which realizes the spatio-temporal distribution prediction of suicide rate in the global range through multi-dimensional data fusion modeling. Specifically, the following steps are included: S1, determination of research scope: based on the data of member states of the World Health Organization, the boundaries of 181 countries and the annual time series from 2000 to 2019 are determined.

[0018] S2, collection of suicide rate data: obtain annual suicide rate data of each country from the World Health Organization Global Health Observatory database, record the data source, statistical method and adjustment coefficient.

[0019] S3, multi-source data acquisition and integration, specifically four types of basic data need to be obtained: S31, natural environment parameters: including precipitation, temperature, extreme high temperature, extreme low temperature, extreme temperature difference, annual average temperature difference, maximum average temperature, minimum average temperature, hottest seasonal temperature, coldest seasonal temperature, NDVI, CO2 emissions, solar radiation; S32, economic parameters: including education investment and GDP proportion, public health investment and GDP proportion, Gini coefficient, per capita GDP, GDP growth rate, unemployment rate, poverty population proportion; S33, social parameters: including male and female ratio, young and old ratio, crime rate, violent conflict index, terrorism index; S34 disease factors: including environmental temperature exposure, autism prevalence, depression prevalence, self-injury rate, anxiety prevalence, schizophrenia prevalence, bipolar prevalence, epilepsy prevalence, malnutrition rate, binge eating disorder prevalence, anorexia prevalence, cancer prevalence, drug and alcohol dependence rate.

[0020] S4, data preprocessing: S41, standardization processing is performed on continuous parameters, and the formula is: x'=(x-μ) / σ, where μ is the mean and σ is the standard deviation; S42, multiple imputation method is used for imputation of missing values, and log transformation is performed on the characteristics of skew distribution.

[0021] S5, after completing all data collection and preprocessing, model training is started, including the following steps: S51, at least linear regression, K nearest neighbor regression, decision tree, XGBoost, multilayer perceptron, and random forest algorithms are used to train regression models; S52, the data set is randomly divided into training samples Xtrain, Ytrain, test samples Xtest, and Ytest according to a certain proportion; S53, the model hyperparameters are optimized, including the number of decision trees, maximum depth, minimum sample splitting, etc., and the determination coefficient (R²) and root mean square error (RMSE) on the test set are used as evaluation indexes to select the optimal model.

[0022] S6, future suicide rate prediction: input the preprocessed influence factor data of the future prediction period into the trained model, and output the predicted suicide rate.

[0023] S7, the results are visualized and analyzed, and a global suicide rate spatial distribution map is generated to represent the suicide rate level with a color gradient; a suicide rate time trend chart of each country is drawn; an influence factor importance ranking chart is generated; annual, five-year, and other different time scale prediction results are supported.

[0024] Please refer to Figure 1The application provides a global suicide rate prediction method based on multi-source data and a machine learning model, which realizes accurate prediction of global suicide rate in combination with a random forest algorithm in view of the deficiencies of existing suicide rate prediction methods in data integration and model accuracy.

[0025] The case area selects countries with complete suicide rate data records and relatively complete multi-source data in the World Health Organization Global Health Observatory Station database. As countries with complete suicide rate data in the world, and natural environment parameters, economic parameters, social parameters and disease parameters in the same year as the suicide rate data can be obtained as research objects, it is convenient for us to study the future suicide rate trend and influencing factors from multiple angles. The research mainly includes the following steps: S1, research scope determination: selecting data of 181 member states of the World Health Organization from 2000 to 2019 as research samples.

[0026] S2, suicide rate data collection: obtaining standardized suicide rate data from the World Health Organization Global Health Observatory Station database, unit: number of suicides per 100,000 people.

[0027] S3, multi-source data acquisition, multi-source data fusion modeling: Natural environment parameters: obtain precipitation, temperature, extreme high temperature, extreme low temperature, extreme temperature difference, annual average temperature difference, maximum average temperature, minimum average temperature, hottest seasonal temperature, coldest seasonal temperature, NDVI, CO2 emission, solar radiation data from the World Meteorological Organization, NASA and other institutions; wherein, the temperature and precipitation data are from the CRU TS climate data set of the University of East Anglia, UK; CO2 data comes from the National Earth System Science Data Center Global 2*2.5 resolution monthly atmospheric carbon dioxide concentration grid simulation data set (1992-2020); other temperature indicators come from WorldClim; solar radiation comes from the solar data set (Global Solar Atlas); NDVI is obtained from the MODIS and GIMMS data sets of NASA; Economic parameters: obtain education investment and GDP proportion, public health investment and GDP proportion, Gini coefficient, per capita GDP, GDP growth rate, unemployment rate, poverty population proportion from the World Bank database; Social parameters: obtain male-female ratio, young-old ratio, crime rate, violent conflict index, terrorism index from the United Nations Development Programme; wherein, the male-female ratio is calculated by the total male and female population of each country in the world from the World Bank; the young-old ratio is also calculated by the population structure data of each country in the world from the World Bank (60 years old or younger divided by 60 years old or older); the crime rate comes from the World Bank; the violent conflict index and the terrorism index come from the Institute for Health Metrics and Evaluation (IHME).

[0028] Disease factors: Environmental temperature exposure, prevalence of autism, depression, self-harm, anxiety disorders, schizophrenia, bipolar disorder, epilepsy, malnutrition, bulimia, anorexia, cancer, and drug and alcohol dependence were obtained from the Global Health Evaluation Institute. Table 1 shows some of the multi-source data. Table 1 shows some of the multi-source data. .

[0029] S4. Preprocessing of the prediction targets and environmental and social factors: The suicide rate was matched with multi-source data using Microsoft Office Excel and Matlab. Data cleaning and integration were performed using Python's pandas library. Continuous parameters were standardized using the formula: x'=(x-μ) / σ, where μ is the mean and σ is the standard deviation. One-hot encoding was performed on categorical variables. Logarithmic transformation was applied to sklearn's IterativeImputer for multiple imputation of missing values.

[0030] S5. Model training uses Matlab software for programming, hyperparameter optimization, and data partitioning training.

[0031] S51. Before model training, the dataset is randomly split into training samples Xtrain and Ytrain, and test samples Xtest and Ytest, in an 8:2 ratio (train:test). Iterative training is then performed, using the coefficient of determination on the test set as a key evaluation metric for stopping iteration. The dataset is re-split in each iteration to prevent data leakage and excessively high coefficients of determination. The table below shows the evaluation results of multiple models: Table 2 Evaluation results of the six models: .

[0032] As shown in Table 2, the random forest model maintained high performance (R²) on the test set. Test =0.9), and compared with the training set performance (R²) Train The closest result is 0.94, indicating its extremely strong generalization ability and good predictive stability for unknown data. In contrast, K-nearest neighbor regression (R²) is the closest. Train =1,R²s Test =0.823) and XGBoost (R² Train =0.98,R² Test= 0.9) performed better on the training set, but their generalization performance was worse than or equal to that of the random forest, and they showed more obvious signs of overfitting. The optimal prediction accuracy and error control: the random forest model achieved the lowest RMSE (1.98) on the test set. RMSE is an absolute measure of the deviation between predicted and true values, and the lower the value, the higher the prediction accuracy. This indicates that the prediction results of the random forest model are the most reliable and have the smallest error, significantly better than other complex models (such as decision trees, MLP, XGBoost, and KNN). Optimal balance between performance and complexity: although the linear regression model did not overfit, its goodness of fit was too low (R² Test <0.3), and it could not capture the complex nonlinear relationships in the data, making it not practical. The random forest model, while ensuring high accuracy, effectively avoids the overfitting risk inherent in weak models such as decision trees through ensemble learning strategies, achieving the optimal balance between prediction performance and model robustness. In summary, the random forest model is proven to be the optimal choice for solving the multi-dimensional prediction problem in this invention due to its excellent generalization ability, lowest prediction error, and best robustness.

[0033] S52、Comprehensive evaluation of the root mean square error of the decision coefficient level of the training set and the test set, and the random forest is selected as the best model, and the hyperparameters of the random forest model include: the number of decision trees is set to 200, and the out-of-bag estimation (OOB) and feature importance calculation functions are enabled. The sample is calculated by calculating the importance of the variable, where M is the number of OOB samples, and and are the predicted value and true value of the m-th OOB sample, respectively. This evaluation method can identify the features that have the greatest impact on the regression task, providing a basis for subsequent feature selection and model optimization.

[0034] The parallel computing function of MATLAB is enabled during model training, and by setting statset('UseParallel', true), multiple decision trees are trained in parallel using multi-core processors, and the risk of overfitting is reduced by limiting the depth of the tree, thereby significantly improving the training efficiency. During the training process, the random forest algorithm completes the regression task by constructing N decision trees. The observation prediction sample distribution graph of the validation set and test set after training with the random forest is as Figure 2 shown.

[0035] The best-performing random forest model in Matlab is saved for subsequent data prediction using the model.

[0036] S6, In addition to predicting the suicide rate of countries around the world, the statistical and data processing methods of the collected influence factors in the future period are the same as S4, and if there is missing future data, use the data in the year with the most recent data to fill in the missing data.

[0037] Subsequently, the trained model saved in S5 was also called using Matlab software, the processed parameters in S6 were brought in, and the model prediction data under the corresponding parameters were exported. Table 3 below shows part of the prediction data: Table 3 Part of the model prediction data .

[0038] S7, visualization analysis and application: S71, using QGIS software to draw a global suicide rate spatial distribution map, color gradient is divided into five quantiles. The time change of suicide rate in the prediction period is visualized. Interactive time trend chart is drawn using Plotly, and the suicide rate change of each country in different periods can be queried.

[0039] S72, according to the importance ranking chart of environmental factors generated by the random forest model, the most important influencing factors can be identified, which has guiding significance for risk control.

[0040] In addition, it should be understood that the above specific embodiments of the present application are only used for illustrative or explanatory purposes of the principles of the present application, and do not constitute a limitation on the present application. Those skilled in the art can make various modifications, replacements and improvements to the present application according to the above content without departing from the spirit and scope of the present application. The technical solutions after these modifications, equivalent replacements and improvements, as long as they meet the technical features defined in the claims of the present application and fall within the equivalent range, should be covered within the protection scope of the present application.

Claims

1.A method for predicting global suicide rate based on multi-dimensional parameters and random forest model, characterized in that, The method comprises the following steps: Step 1: determining the list of countries and time series included in the research scope; Step 2: collecting the suicide rate data of each country year by year as the target variable; Step 3: constructing a multi-dimensional prediction index system, obtaining the modeling basis data of each country in the research scope, using the random forest regression model in the machine learning library, selecting the optimal model parameters through model training and verification, and updating the modeling basis data to realize the updateable global suicide rate prediction model; Step 4: predicting the future suicide rate of the target country; Step 5: visualizing the spatial distribution map of the global suicide rate prediction results and the suicide rate trend chart of each country. 2.The global suicide rate prediction method based on multi-dimensional parameters and random forest model according to claim 1, wherein, In step 2, the suicide rate data of each country is labeled with data source, statistical method and adjustment coefficient to ensure data comparability. 3.The global suicide rate prediction method based on multi-dimensional parameters and random forest model according to claim 1, wherein, In step 3, the modeling basis data includes natural environment parameters, economic parameters, social parameters and disease factors; based on the acquisition of multi-variable modeling basis data, parameter interaction items are introduced when constructing the multi-dimensional prediction index system to enhance the explanation depth and prediction accuracy of the model. The interaction item reveals the action mechanism reflected by multiple variables by capturing the synergistic or antagonistic effect between variables. 4.The global suicide rate prediction method based on multi-dimensional parameters and random forest model according to claim 1, wherein, In step 3, the specific implementation process of the global suicide rate prediction model is as follows: 3.1) Select a machine learning model: select a regression model suitable for multi-factor prediction of suicide rate from the machine learning library; 3.2) Data preprocessing: convert parameters of different dimensions into comparable values through standardization processing; 3.3) Establish a data set: the input data set includes feature values X and target values Y, the target value Y represents the suicide rate of a country in a year, and the feature value X represents the multi-dimensional factors affecting Y; 3.4) Sample segmentation: divide the data set into training samples Xtrain, Ytrain and test samples Xtest, Ytest in proportion; the training samples are used to learn the rules of historical data; the test samples are used to evaluate the prediction performance of the model; when dividing, the data integrity of a country is maintained, that is, all the data of a country in all years are used as the training set or the test set, which ensures randomness and improves the generalization ability of the model; 3.5) Train the model: set different hyperparameter combinations, input the training samples Xtrain and Ytrain into the model under different combination conditions; at the same time, input Xtrain of the training sample and Xtest of the test sample into the model, compare Ypre_train and Ypre_test predicted by the model with Ytrain and Ytest respectively, take the coefficient of determination and the root mean square error as the evaluation parameters, and select the model with the smallest error and the hyperparameters. 5.The global suicide rate prediction method based on multi-dimensional parameters and random forest model according to claim 4, characterized in that, Step 3.2) is as follows: Standardize continuous parameters, one-hot encode categorical variables, and logarithmically transform features with skewed distribution. 6.The global suicide rate prediction method based on multi-dimensional parameters and random forest model according to claim 1, wherein, In step 5, the spatial distribution map of the global suicide rate prediction results in the same period is color-coded according to different suicide rate levels; the suicide rate trend chart of the same country in different periods presents the time trend through a line chart.

Images