OEFA: An Optimized Ensemble Fusion Algorithm for Stress Classification and Prediction

Abstract

Stress is a prevalent issue in modern society, leading to various adverse effects on individuals' mental and physical well-being. Accurate classification and prediction of stress levels are crucial for effective intervention and support. However, existing stress classification and prediction techniques suffer from several disadvantages, including imbalanced datasets, missing values, categorical data, lack of feature selection, and limited model diversity. To address these challenges, this paper proposes an Optimized Ensemble Fusion Algorithm (OEFA) for stress classification and prediction. The OEFA algorithm combines multiple techniques to overcome the limitations of existing approaches. It employs the Cluster-Based Adaptive Synthetic Sampling (CBADAS) algorithm to balance imbalanced datasets, generating synthetic instances for the minority class. Missing values are handled by removing instances with missing data, ensuring only complete data is used for training and testing. Categorical data is transformed into numerical format using label encoding, enabling the use of traditional machine learning algorithms. Attribute selection is performed using the ReliefFAttributeEval algorithm with Ranker Search, reducing dimensionality and improving computational efficiency. Furthermore, OEFA leverages ensemble learning with the AdaBoostM1 algorithm, incorporating optimized versions of Logistic Regression, HoeffdingTree, LMT, REPTree, JRip, OneR, PART, and MultilayerPerceptron as base classifiers. Experimental results demonstrate the superiority of the OEFA algorithm, exhibiting the highest accuracy, precision, recall, and F1-score compared to existing techniques. The advantages of OEFA include improved accuracy through addressing imbalanced datasets, missing values, and categorical data. The algorithm enhances generalization capability, mitigates overfitting, and demonstrates robustness by combining multiple base classifiers. Efficient feature selection is achieved using ReliefFAttributeEval with Ranker Search, contributing to reduced dimensionality and computational requirements.