Estimating Students' Adjustment Level in Distance Education Using Machine Learning and Resampling

Abstract

COVID-19's impact catalyzed the integration of distance education into our lives, shaping a crucial facet of modern learning. Initially, students, educators, and administrators encountered diverse challenges in navigating this paradigm shift. Swiftly addressing these hurdles promises enhanced efficacy in remote education. This research, blending experimental and descriptive methodologies, scrutinizes the "Students Adaptability Level in Online Education" dataset. It aims to assess students' adaptability in distance learning using five distinct machine learning techniques and identify pivotal factors influencing adaptation. Multiple classification endeavors aim to bolster predictive accuracy. Leveraging 14 resampling approaches, 70 classifications per algorithm—both with and without sampling—were conducted, each meticulously evaluated using four performance metrics. The Random Forest model, coupled with KMeansSMOTE oversampling, yielded a notable 93% accuracy, showcasing heightened classifier efficacy through resampling. Noteworthy correlations emerged, indicating that lesson durations of 1-3 hours, reliable internet connectivity, and financial assistance to families correlate with enhanced student adaptation. This study underscores the potential of resampling techniques in refining classification accuracy and underscores actionable strategies for optimizing distance education's effectiveness.