Enhancing Diagnostic Decision-Making: Ensemble Learning Techniques for Reliable Stress Level Classification.

Area of Science:

• Computer Science
• Psychology
• Education

Background:

• Intense academic pressure leads to student stress, with potential adverse mental and physical effects.
• The shift to online learning has increased student workloads and stress levels.
• Students may exhibit stress through sleep deprivation and altered eating habits.

Purpose of the Study:

• To propose a novel ensemble learning architecture for classifying student stress levels.
• To identify key factors contributing to academic stress, including sleep, screen time, and study habits.
• To develop an effective model for predicting and managing student academic stress.

Main Methods:

• Collected survey data from students on sleep hours, productive time, screen time, assignments, and study methods.
• Preprocessed data to categorize stress into 'highly stressed,' 'manageable stress,' and 'no stress.'
• Employed oversampling for minority class imbalance and implemented ensemble learning algorithms (Decision Tree, Random Forest, AdaBoost, Gradient Boost).

Main Results:

• The ensemble learning model achieved 93.48% accuracy and 93.14% F1 score.
• Fivefold cross-validation yielded an accuracy of 93.45%.
• Receiver Operating Characteristic (ROC) analysis showed 98% accuracy for 'no stress' and 91% true positive rate for 'manageable' and 'high stress' categories.

Conclusions:

• The proposed ensemble learning approach with cross-validation accurately predicts student stress levels, outperforming state-of-the-art algorithms.
• This model can help students identify areas for improvement and reduce academic stress.
• The findings support the development of effective stress prediction tools to improve academic lifestyles.