Mental health classification method for army officers and soldiers

Abstract

The invention discloses an army officer and soldier mental health classification method in the technical field of mental health, and the method comprises the following specific steps: 1, obtaining psychological values A of a plurality of samples and a plurality of sample feature value matrixes I, the psychological values A being 0 or 1; step 2, storing a plurality of psychological values A and a plurality of sample characteristic value matrixes I in a table form; step 3, carrying out standardization processing on the sample characteristic value matrix I after the label listing, extracting principal component characteristics based on principal component analysis, and then selecting characteristic vectors corresponding to the first k maximum characteristic values to form a transformation matrix; and step 4, performing model training for the dichotomy task, finally predicting principal component features of the test set, and outputting a probability value or a category label of the mental health state. According to the method, through combination of principal component analysis and machine learning, high-dimensional feature dimension reduction and automatic risk prediction are realized, and the method is suitable for mental health classification in army officer and soldier training, resident training and task deployment scenes.

Description

Technical Field

[0001] This invention relates to the field of mental health technology, specifically to a method for classifying the mental health of military personnel. Background Technology

[0002] For officers and soldiers in the armed forces, they need to undergo high-intensity military training for a long time to cope with possible combat missions. Such high-intensity training and combat missions are often accompanied by enormous psychological pressure, such as physical fatigue, mental tension, and life-threatening situations. Being in such a high-pressure environment for a long time can easily lead to psychological problems such as anxiety, depression, and fear among officers and soldiers.

[0003] The assessment of military personnel's mental health has long relied on standardized scales and manual interviews, which has significant limitations. Assessment results are influenced by the subjects' expressed intentions and cognitive biases, and are particularly prone to missing latent psychological risks during high-intensity tasks. Furthermore, relying solely on textual self-reports lacks objective evidence such as physiological indicators and behavioral data, making it difficult to predict changes in soldiers' mental health. Simultaneously, existing methods have vague data structure definitions and unclear logical relationships between labels and features, leading to difficulties in cross-modal data fusion. Examples include methods used in military personnel health checkups, the seven dimensions of military personnel's mental health, the new recruit mental health information collection system, and the basic occupational adaptability test for military academy recruits.

[0004] Therefore, this invention proposes a method for classifying the mental health of military personnel, which is applicable to the mental health classification of military personnel in training, field training, and mission deployment scenarios. By combining principal component analysis with machine learning, it achieves dimensionality reduction of high-dimensional features and automated risk prediction. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method for classifying the mental health of military personnel. By combining principal component analysis with machine learning, it achieves dimensionality reduction of high-dimensional features and automated risk prediction, making it applicable to the mental health classification of military personnel in training, field training, and mission deployment scenarios.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A method for classifying the mental health of military personnel, comprising the following specific steps:

[0007] Step 1: Obtain the psychological values ​​A of several samples and the feature value matrix I of several samples. The psychological value A is 0 or 1, and the psychological value A represents the mental health status of the sample. When the psychological value A of a sample is 0, the psychological condition of the sample is defined as mental health, and when the psychological value A of a sample is 1, the psychological condition of the sample is defined as psychological danger.

[0008] I∈}a1,a2,…,a n};

[0009] Where a1, a2, ..., a n These represent different feature dimensions, where n is a constant.

[0010] Step 2: Store several psychological values ​​A and several sample feature value matrices I in tabular form; set psychological value A as a label column X, and all columns after label column X are sample feature value matrices I, with each column corresponding to a feature dimension and each row corresponding to a complete feature vector of a sample.

[0011] Step 3: Standardize the sample feature value matrix I after the label column, extract principal component features based on principal component analysis, and then select the feature vectors corresponding to the first k largest feature values ​​to form the transformation matrix.

[0012] Step four: Based on the binary classification task, the random forest algorithm is selected, and the dataset is divided into training and test sets in a 7:3 ratio. Then, 5-fold cross-validation is used to train the model on the training set. The hyperparameters are optimized through grid search to obtain the prediction model. Finally, the trained prediction model is used to predict the principal component features of the test set and output the probability value or category label of mental health status.

[0013] Furthermore, the psychological value A in step one can be obtained through: manual rating results based on standardized scales, professional judgments given by psychologists based on interviews or observations, and historical classification records.

[0014] Furthermore, in step one, the source of several sample feature values ​​I is the structured record data of military personnel, which includes physiological indicators, behavioral data, and test scores.

[0015] Furthermore, in step three, missing values ​​in the feature data are processed using one of the following methods: mean imputation, median imputation, or multiple imputation.

[0016] Furthermore, the principal component feature extraction process in step three includes: calculating the standardized sample feature matrix, obtaining the corresponding eigenvalues ​​and eigenvectors through singular value decomposition, and sorting them from largest to smallest according to the eigenvalues.

[0017] Furthermore, the process of selecting the value of k in step three includes: determining the number of principal components k by setting a preset cumulative contribution rate threshold or scree plot, and retaining principal components with a cumulative contribution rate ≥ 85%.

[0018] Furthermore, the process of constructing the principal component feature matrix in step three is as follows: the transformation matrix W is constructed using the first k eigenvectors, and the original feature matrix I is transformed into the principal component feature matrix Z using the formula Z = I·W.

[0019] Furthermore, step five involves integrating the trained prediction model into the military's mental health management system, connecting to a real-time data interface, calculating the principal component feature matrix of new samples in real time, inputting it into the model to automatically output the mental state prediction results, and assisting the mental health department in formulating intervention strategies.

[0020] Furthermore, in step four, the model performance is evaluated using accuracy, precision, recall, F1 score, and ROC-AUC curve, and the recall rate of risky samples is optimized.

[0021] Furthermore, the standardization process includes at least one of Z-score standardization and Min-Max normalization.

[0022] The above-mentioned approach has the following beneficial effects: 1. This invention, through the organic combination of principal component analysis and machine learning, constructs a scientific and efficient technical system for classifying the mental health of military personnel, fundamentally breaking through the bottleneck of traditional psychological assessment methods in high-dimensional data processing. In complex scenarios such as military training, field training, and mission deployment, the mental health status of military personnel is affected by multiple dimensions such as physiological indicators, behavioral data, and test scores. The original data often exhibits high dimensionality and strong correlation. Principal component analysis technology can transform the complex feature space into a few comprehensive principal components by feature dimensionality reduction, while retaining key information with a cumulative contribution rate of ≥85%. This significantly reduces the complexity of data computation and effectively avoids the negative impact of the curse of dimensionality on model performance. Meanwhile, the random forest algorithm, as a strong classifier, has the advantages of handling nonlinear relationships and suppressing overfitting. The synergistic effect of the two algorithms achieves full-process optimization from data preprocessing to automated risk prediction, providing stable and reliable technical support for mental health assessment in dynamic scenarios, enabling the military to promptly capture subtle changes in the mental state of military personnel during high-intensity missions.

[0023] This method establishes a standardized and regulated data processing workflow, significantly improving the accuracy and practicality of mental health classification. From the data acquisition stage, it integrates multiple sources of psychological values, including standardized scale scores, physician professional judgment, and historical classification records, combined with structured records of physiological behaviors, ensuring the comprehensiveness and authority of the basic data. In the data preprocessing stage, Z-score standardization and missing value imputation techniques are used to clean and optimize the original features, laying a high-quality data foundation for subsequent modeling. Particularly during model training, a 7:3 ratio is used to divide the training and test sets, combined with 5-fold cross-validation and grid search hyperparameter optimization strategies, effectively uncovering potential correlations between data features and balancing model bias and variance. The design, which focuses on optimizing the recall rate of risk samples for binary classification tasks, aligns with the actual needs of military mental health management, minimizing missed diagnoses of psychologically vulnerable individuals and providing a window of opportunity for early intervention, demonstrating a deep fit between the technical solution and military application scenarios.

[0024] 2. This invention establishes a complete closed loop from data processing to practical application, providing an innovative solution for the intelligent transformation of military mental health management. By integrating the trained predictive model into the mental health management system and connecting it to a real-time data interface, the entire process from data collection and feature transformation to risk prediction is automated. This feature enables the system to respond to new sample inputs in real time and quickly output psychological state prediction results, providing accurate decision-making basis for mental health departments to formulate personalized intervention strategies. In mission scenarios with high mobility and changing environments, such as field training and exercises, this system can overcome the time and space limitations of manual assessment, dynamically monitor and warn of the mental state of officers and soldiers, and help build a full-chain management system of "prevention-identification-intervention". At the same time, the application of principal component analysis technology retains the core features that have the most significant impact on mental health, making it easier for mental health workers to interpret key influencing factors from the reduced feature space. While improving management efficiency, it provides scientific guidance for the design of targeted psychological intervention programs, promoting the development of military mental health support work towards precision and intelligence.

[0025] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an embodiment of the method for classifying the mental health of military personnel according to the present invention. Detailed Implementation

[0027] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The following detailed description illustrates the specific implementation method:

[0029] Example:

[0030] As attached Figure 1 As shown, a method for classifying the mental health of military personnel includes the following specific steps:

[0031] Step 1: Obtain psychological values ​​A and feature value matrix I for several samples. Psychological values ​​A are either 0 or 1, representing the mental health status of the sample. When a sample's psychological value A is 0, the sample's psychological condition is defined as mentally healthy; when a sample's psychological value A is 1, the sample's psychological condition is defined as psychologically dangerous. The psychological values ​​A are obtained through methods including: manual rating results based on standardized scales, professional judgments given by psychologists based on interviews or observations, and historical classification records. The feature values ​​I are obtained from structured record data of military personnel, which includes physiological indicators, behavioral data, and test scores.

[0032] I∈{a1, a2, ..., a n};

[0033] Where a1, a2, ..., a n These represent different feature dimensions, where n is a constant.

[0034] Step 2: Store several psychological values ​​A and several sample feature value matrices I in tabular form; set psychological value A as a label column X, and all columns after label column X are sample feature value matrices I, with each column corresponding to a feature dimension and each row corresponding to a complete feature vector of a sample.

[0035] Step 3: Standardize the sample feature value matrix I following the label column (Z-score standardization or Min-Max normalization), then extract principal component features based on principal component analysis, and finally select the eigenvectors corresponding to the k largest eigenvalues ​​to form the transformation matrix. For missing values ​​in the feature value data, one of the following methods is used: mean imputation, median imputation, or multiple imputation.

[0036] The principal component feature extraction process includes: calculating the standardized sample feature matrix, obtaining the corresponding feature values ​​and feature vectors through singular value decomposition, and sorting the feature values ​​from largest to smallest. The selection of the k value is determined by a preset cumulative contribution rate threshold or scree plot to determine the number of principal components k, and retaining principal components with a cumulative contribution rate ≥ 85%.

[0037] The process of constructing the principal component feature matrix is ​​as follows: the transformation matrix W is constructed using the first k eigenvectors, and the original feature matrix I is transformed into the principal component feature matrix Z using the formula Z = I·W.

[0038] Step four: Based on the binary classification task, select the random forest algorithm (in actual algorithm selection, consider the task objectives (such as prediction accuracy, interpretability, deployment efficiency), data characteristics (sample size, feature complexity, class balance), and the needs of the military application scenario, and select suitable solutions from a variety of mature binary classification algorithms, such as gradient boosting trees and logistic regression). Divide the dataset into training and test sets in a 7:3 ratio. Then, use 5-fold cross-validation to train the model on the training set, optimize the hyperparameters through grid search, obtain the prediction model, and finally use the trained prediction model to predict the principal component features of the test set, outputting the probability value or category label of mental health status.

[0039] The model performance was evaluated using accuracy, precision, recall, F1 score, and ROC-AUC curves, and the recall rate for risky samples was optimized.

[0040] In addition, step five involves integrating the trained prediction model into the military mental health management system, connecting to the real-time data interface, calculating the principal component feature matrix of new samples in real time, inputting the model to automatically output the mental state prediction results, and assisting the mental health department in formulating intervention strategies.

[0041] The specific implementation is as follows: Taking a mechanized infantry brigade of a certain army group as the application object, the method was implemented to classify the psychological health needs of officers and soldiers during the training period, and the effectiveness and applicability of the method were verified. The brigade has three combined arms battalions with a total of 1,200 officers and soldiers. 800 officers and soldiers who participated in the mission during the 2023 training cycle (90 days) were selected as the sample group, covering different ranks (private to major), branches of service (infantry, armored corps, communications corps) and mission positions (combatant, support officer, commander) to construct a representative dataset.

[0042] I. Data Acquisition and Preprocessing (Steps 1 and 2):

[0043] 1. Psychological value A is obtained by collecting psychological state labels using a multi-source fusion method:

[0044] Standardized scale scoring: Monthly assessments were conducted using the Military Mental Health Scale (MHS), the PHQ-9 Depression Scale, and the GAD-7 Anxiety Scale. The psychologists of the medical unit assigned a psychological score (A) based on the scoring criteria. For example, an MHS total score <70, a PHQ-9 score <10, and a GAD-7 score <8 were considered psychologically healthy (A=0); conversely, a score above these thresholds, combined with clinical assessment, was marked as psychologically at risk (A=1).

[0045] Professional interview assessment: A round of face-to-face interviews is conducted every 30 days during the training period. Psychologists observe the soldiers' emotional state, focus on task execution, frequency of social interaction and other behavioral characteristics, and make risk assessments based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) standards, and revise the scale scores accordingly.

[0046] Historical classification records: Historical data such as the psychological crisis intervention records of officers and soldiers of the brigade and the results of psychological screenings before leave were integrated in the past year to form a cross-cycle psychological state label library. Finally, 782 valid psychological value label samples were obtained, of which 615 samples were A=0 (mental health) and 167 samples were A=1 (psychological risk). The label distribution is consistent with the actual proportion of psychological risk in the troops.

[0047] 2. Construction of Sample Feature Value Matrix I: Collection of 18-dimensional structured feature data across 3 major categories:

[0048] Physiological indicators: Resting heart rate, sleep duration, heart rate variability (RMSSD), and average daily steps are collected in real time via smart bracelet; blood pressure, blood oxygen saturation, and cortisol concentration (morning detection value) are obtained via medical terminal.

[0049] Behavioral data: The training management system extracts the average daily training time, the fluctuation value of live-fire shooting hit rate, and the number of tactical coordination errors; the attendance system records the number of late roll call attendances and the frequency of participation in group activities; the communication log statistics the daily mobile phone usage time (distinguishing between work / entertainment purposes).

[0050] The assessment scores are as follows: military training assessment score (including comprehensive scores of physical fitness, skills, and tactics), mission attitude questionnaire score (subjective initiative and stress coping strategies), and peer evaluation score (teamwork and emotional stability indicators). Taking a private as an example, the original feature vector is: [20 years old, 3 years of military service, infantry, resting heart rate 72 beats / min, sleep 5.8 hours, RMSSD 45ms, average daily steps 12,000 steps, blood pressure 120 / 80 mmHg, blood oxygen 98%, cortisol 120 nmol / L, training duration 8.5 hours, hit rate fluctuation 3%, number of errors 2, lateness 0, participation in group activities 4 times / week, mobile phone use 3.2 hours (1.5 hours for entertainment), training performance 85 points, mission attitude 92 points, peer evaluation 88 points].

[0051] The data was stored in an Excel spreadsheet. The first column was the label column X (psychological value A), and the following 18 columns were the feature dimensions, named as follows: age, years of service, branch of service (0 = infantry, 1 = armored, 2 = communications), resting heart rate, sleep duration, RMSSD, average daily steps, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, cortisol, training duration, hit rate fluctuation, number of errors, number of late arrivals, frequency of group activities, mobile phone usage time, entertainment usage ratio, training performance, mission attitude, and peer evaluation. Each row corresponds to a complete feature vector of a soldier, generating a total of 782 rows of data records.

[0052] II. Feature Reduction and Principal Component Analysis (Step 3):

[0053] 1. Data cleaning and standardization

[0054] Missing value handling: For features such as sleep duration and cortisol that have 5% missing values, multiple imputation (MICE) is used to construct a regression model based on relevant features such as age, length of service, and branch of service to predict missing values, avoiding information bias caused by single mean imputation.

[0055] Standardization: Z-score standardization is used. Taking training time as an example, the calculation formula is as follows:

[0056]

[0057] Where μ is the mean of the feature (7.2 hours) and σ is the standard deviation (1.5 hours), all features are scaled to a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the influence of dimensions.

[0058] 2. Implementation of Principal Component Analysis

[0059] Covariance matrix calculation: The covariance matrix is ​​calculated on the standardized 18-dimensional feature matrix to identify linear correlations between features. For example, it was found that the correlation coefficient between sleep duration and RMSSD is 0.78, and the correlation coefficient between training performance and task attitude is 0.82, indicating significant multicollinearity and verifying the necessity of dimensionality reduction.

[0060] Singular Value Decomposition (SVD): The PCA module is called using the scikit-learn library in Python to calculate eigenvalues ​​and eigenvectors. The eigenvalues ​​are sorted from largest to smallest. The top 6 eigenvalues ​​are 5.2, 3.8, 2.9, 2.1, 1.8, and 1.3, with a cumulative contribution rate of 88.7% (exceeding the 85% threshold). Finally, k=6 is determined.

[0061] Scrap plot verification: A line graph of eigenvalues ​​and principal component numbers was plotted. The eigenvalues ​​of the first 6 principal components showed a significant decreasing trend, while those of the 7th and subsequent eigenvalues ​​tended to level off, which is consistent with the cumulative contribution rate analysis results. This confirms that the first 6 principal components should be retained.

[0062] Principal component feature matrix construction: Extract the first 6 feature vectors to form a 6×18 dimensional transformation matrix W, and then use matrix multiplication Z = I×WT to transform the original 18-dimensional features into 6-dimensional principal component features. For example, the principal component feature vector of a sample is: [2.1,-1.3,0.8,1.5,-0.9,0.6], which contains 88.7% of the key information of the original data.

[0063] III. Model Training and Evaluation (Step Four)

[0064] 1. Dataset partitioning and model configuration

[0065] Data partitioning: The training set (547 samples) and the test set (235 samples) were randomly divided in a 7:3 ratio. Stratified sampling was used to ensure a balanced ratio of the two types of samples (117 psychologically risky samples in the training set and 50 samples in the test set).

[0066] Random forest parameter settings: initial parameters are n_estimators=100, max_depth=None, min_samples_split=2. Five-fold cross-validation is used, and the training set is further divided into 5 subsets. Each time, 4 subsets are used for training and 1 subset is used for validation. The average performance index is taken after 5 iterations.

[0067] Grid search optimization: A grid search was performed for n_estimators(50-200, step size 25), max_depth(5-20, step size 5), and min_samples_leaf(1-5, step size 1). The F1 score and recall were calculated for each parameter combination. The optimal parameter combination was n_estimators = 150, max_depth = 15, and min_samples_leaf = 2.

[0068] 2. Model Training and Performance Evaluation

[0069] Training process: The model is trained using optimized parameters. Each decision tree is split by randomly selecting 80% of the training samples and 3 dimensions of the 6 principal components based on bootstrap sampling to suppress overfitting. Training takes approximately 12 minutes.

[0070] Finally, by using the principal component weights output by the model to infer the importance of the original features, it was found that sleep duration (corresponding to principal component 1 weight 0.65), RMSSD (0.58), and task attitude score (0.52) are the top three features affecting mental health, providing targets for psychological intervention.

[0071] A specific experiment was conducted using the aforementioned classification method for the mental health of military personnel:

[0072] I. Experimental Subjects

[0073] Experimental unit: A mechanized infantry brigade of a certain group army. Total number of officers and soldiers: 1200. Sample selection: 800 officers and soldiers who participated in the mission during the 2023 training cycle (90 days).

[0074] II. Experimental Data Tables

[0075] Table 1 - Core Indicators for Model Training and Evaluation

[0076]

[0077]

[0078] As shown in Table 1, the model training used a 7:3 dataset split, with stratified sampling to ensure a balanced distribution of psychologically risky samples (117 training samples and 50 test samples). Through 5-fold cross-validation and grid search optimization, the optimal parameters for the random forest were determined to be n_estimators = 150 and max_depth = 15, improving generalization ability while maintaining computational efficiency. Test set evaluation showed that the model achieved a recall rate of 88% for the psychologically risky class, significantly higher than traditional methods (approximately 70%), indicating its effective capture of high-risk individuals. Accuracy of 92.3%, F1 score of 85.9%, and ROC-AUC area of ​​0.94 demonstrate excellent classification performance and good robustness. Feature importance analysis revealed that sleep duration, RMSSD (heart rate variability), and task attitude scores were the core influencing factors, providing clear physiological and cognitive intervention targets for psychological intervention. This validated the scientific validity and practicality of combining principal component analysis and machine learning in high-dimensional feature modeling.

[0079] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for classifying the mental health of military personnel, characterized by, The specific steps include the following: Step one, obtain a number of sample psychological values A and a number of sample feature value matrices I, the psychological value A is 0 or 1, the psychological value A represents the mental health status of the sample; wherein when the psychological value A of a certain sample is 0, the psychological condition of the sample is defined as mental health, and when the psychological value A of a certain sample is 1, the psychological condition of the sample is defined as mental risk; I e {a1, a2,..., a n}; wherein a1, a2,..., a n denote different feature dimensions, n is a constant; Step two, store the number of psychological values A and the number of sample feature value matrices I in table form; set the psychological value A as a label column X, and all columns after the label column X are sample feature value matrices I, and each column corresponds to a feature dimension, and each row corresponds to a complete feature vector of a sample; Step three, standardize the sample feature value matrix I after the label column, then extract principal component features based on principal component analysis, and then select the feature vectors corresponding to the first k largest eigenvalues to form a transformation matrix; Step four, based on the binary classification task, select the random forest algorithm, divide the data set into a training set and a test set in a ratio of 7:3; then use 5-fold cross-validation to train the model on the training set, optimize the hyperparameters through grid search, obtain the prediction model, and finally use the trained prediction model to predict the principal component features of the test set, output the probability value or category label of the mental health status.

2. The method of claim 1, wherein: The acquisition approach of the psychological value A in step one includes: manual scoring results based on standardized scales, professional judgments given by psychologists according to interviews or observations, and historical classification records.

3. The method of claim 1, wherein: The source of the sample feature value I in step one is the structured record data of the officers and soldiers, which includes physiological indicators, behavior data and evaluation scores.

4. The method of claim 3, wherein: In step three, the missing values in the feature value data are processed by using one of the mean filling, median filling or multiple imputation method.

5. The method of claim 4, wherein: The principal component feature extraction process in step three includes: calculating the standardized sample feature matrix, then obtaining the corresponding eigenvalues and corresponding eigenvectors through singular value decomposition, and sorting the eigenvalues from large to small.

6. The method of claim 5, wherein: The selection process of the value k in step three includes: determining the number of principal components k by a pre-set cumulative contribution rate threshold or a gravel chart, and retaining the principal components with a cumulative contribution rate of ≥85%.

7. The method of claim 6, wherein: The construction process of the principal component feature matrix in step three is as follows: use the first k eigenvectors to form a transformation matrix W, and convert the original feature matrix I to a principal component feature matrix Z through the formula Z = I·W.

8. The method of claim 7, wherein: Step five, integrate the trained prediction model into the psychological health management system of the army, connect the real-time data interface, calculate the principal component feature matrix of the new sample in real time, input the model to automatically output the psychological state prediction result, and assist the psychological department in formulating intervention strategies.

9. The method of claim 8, wherein: In step four, the model performance is evaluated by accuracy, precision, recall, F1 score and ROC-AUC curve, and the recall rate of the risk sample is optimized.

10. The method of claim 9, wherein: The standardization process at least includes one of Z-score standardization and Min-Max normalization.

Images