THE PSYCHOLOGICAL IMPACT OF COVID-19 ON PEOPLE DURING LOCKDOWN AND ITS EFFECT ON STUDENT EDUCATION: AN ENSEMBLE LEARNING APPROACH

Abstract

The COVID-19 pandemic has had a profound impact on various aspects of life, particularly on mental health and education. This study explores the psychological effects of COVID-19 lockdowns on individuals and the educational disruptions faced by students. This research proposes an advanced ensemble learning approach to analyze two datasets: the "Psychological Impact of COVID-19 on People" and the "Effect of COVID-19 on Student Education." The goal is to accurately predict psychological outcomes such as depression, anxiety, and stress, as well as the educational impact on students. The proposed methodology involves several key steps: data preprocessing, feature scaling, handling class imbalances, feature selection using Recursive Feature Elimination with Cross-Validation (RFECV), and model building using ensemble learning techniques.

Initially, categorical variables such as gender and student status are encoded numerically, and missing values are imputed using the mean of the respective columns. The features are then standardized to ensure uniformity. RFECV is applied to select the most significant features, enhancing the model's performance by eliminating less important features. To handle class imbalances in the data, we utilize the Synthetic Minority Over-sampling Technique (SMOTE), dynamically adjusting the k_neighbors parameter to ensure valid configurations. This step is crucial for creating a balanced dataset that enhances model performance. The model employs three primary classifiers: Random Forest, Support Vector Machine (SVM), and Gradient Boosting. Each classifier undergoes hyperparameter tuning using GridSearchCV, optimizing for accuracy. The best models from each classifier are then combined using a Stacking Classifier, which leverages the strengths of all three classifiers to improve prediction accuracy. The model is trained and evaluated on two datasets. For the psychological impact dataset, the targets are depression, anxiety, and stress. For the student education dataset, the outcome variable is predicted. The ensemble model's performance is assessed using metrics such as precision, recall, F1-score, and accuracy. The proposed model obtained an accuracy of 98.33% in predicting the psychological impact of COVID-19 on people, and 98.46% in predicting the impact of COVID-19 on student education. The results demonstrate the significant improvements of the ensemble model in prediction accuracy, achieving robust performance metrics.