What question did this study set out to answer?

To create a robust risk assessment framework for predicting mental health issues in students using advanced data analytics.

March 18, 2026Open Access

Construction of a Comprehensive Risk Assessment Framework for Mental Health Education Based on Big Data Prediction and Neural Networks

Key Points

To create a robust risk assessment framework for predicting mental health issues in students using advanced data analytics.
Developed a framework combining PCA, Z-score normalization, and SSL-DERNN.
Tuned model hyperparameters with SSLO metaheuristic.
Used stratified 5-fold cross-validation for model evaluation.
Analyzed performance metrics including accuracy, precision, recall, and F1-score.
Achieved accuracy of 97.0% with a 95% confidence interval of 96.1-97.9%.
Precision was 95.5% and recall was 93.2%, both within 95% confidence intervals.
Outperformed several baselines including ANN, RF, and LSTM models (p < 0.01).

Abstract

A risk classification framework for student mental health is proposed, combining Principal Component Analysis (PCA), Z-score normalization, and a Seven-Spot Ladybird-Tuned Dynamic Elman Recurrent Neural Network (SSL-DERNN). Model hyperparameters were tuned using a Seven-Spot Ladybird Optimization (SSLO) metaheuristic. The proposed framework combines the big data analytics with a Seven-Spot Ladybird-Tuned Dynamic Elman Recurrent Neural Network (SSL-DERNN) to handle big, heterogeneous educational and psychological data effectively. The combination of meta-heuristic tuning and dynamic recurrent learning allows learning features robustly and predicting mental-health risks over large groups of students. Experiments used stratified 5-fold cross-validation; reported metrics are mean ±95% confidence intervals across folds. SSL-DERNN achieved accuracy 97.0% ( 95% CI: 96.1-97.9%), precision 95.5% ( 95% CI: 94.2 − 96.8% ), recall 93.2% ( 95% CI: 91.6 − 94.8%) and F1-score 92.6%(95% CI: 91.0-94.2%), significantly outperforming ANN, RF, RF-ANN, LSTM, GRU and BiLSTM baselines (paired tests, p < 0.01 ). Ablation analysis confirmed contributions of PCA and SSLO. Hardware, software, and reproducible hyperparameters are reported to aid replication.

Read Full Paperexternally

Bookmark

View Full Paper

Cite This Study

Deng et al. (Sun,) studied this question.

synapsesocial.com/papers/69ba43884e9516ffd37a4e6b https://doi.org/https://doi.org/10.6180/jase.202608_31.048

Bookmark

View Full Paper