Prediction of college student psychological state based on deep learning framework combining the improved Whale...

Area of Science:

• Artificial Intelligence in Education
• Machine Learning for Mental Health
• Computational Psychology

Background:

• Accurate prediction of college students' psychological states is vital for improving teaching quality and implementing mental health interventions.
• Traditional machine learning methods struggle with the temporal dependency and nonlinearity of psychological data.
• Long Short-Term Memory (LSTM) networks are susceptible to hyperparameter sensitivity and premature convergence.

Purpose of the Study:

• To propose a novel framework, CNWOA-LSTM, that combines an improved Whale Optimization Algorithm (WOA) with LSTM for enhanced psychological state prediction in college students.
• To address the limitations of traditional machine learning and standard LSTM models in handling complex psychological data.
• To optimize key LSTM hyperparameters using an improved WOA for superior predictive performance.

Main Methods:

• Developed the CNWOA-LSTM framework integrating an improved WOA with LSTM.
• Incorporated chaotic initialization in WOA to boost population diversity and global search capabilities.
• Added a niche operator to the WOA to mitigate premature convergence issues.
• Optimized critical LSTM hyperparameters including hidden units, learning rate, dropout rate, and batch size.
• Validated the framework on a public Kaggle dataset containing student behavioral, learning interaction, and psychological assessment data.

Main Results:

• The CNWOA-LSTM framework achieved an accuracy of 93.64% in predicting college students' psychological states.
• Demonstrated significant accuracy improvements of 9.33% over the original LSTM and 6.63% over the standard WOA-LSTM.
• Outperformed other meta-heuristic optimized LSTM models (DE-LSTM, HHO-LSTM, MPA-LSTM, AOA-LSTM, RSA-LSTM) by 5.21%-7.34%.
• Effectiveness was verified through comprehensive comparisons with traditional machine learning and other deep learning models.

Conclusions:

• The proposed CNWOA-LSTM framework effectively predicts college students' psychological states, outperforming existing methods.
• The integration of chaotic initialization and a niche operator in WOA enhances LSTM's performance for complex psychological data.
• CNWOA-LSTM offers a promising approach for advancing intelligent education systems and supporting student mental well-being.