CLASSIFICATION OF UNDERGRADUATE STUDENTS’ ADAPTABILITY LEVELS IN HYBRID LEARNING USING MACHINE LEARNING TECHNIQUES

Abstract

This research investigated the adaptability levels of undergraduate students in hybrid learning environments using machine learning techniques. This research study was aimed to identify crucial factors influencing adaptability and assess the effectiveness of three distinct algorithms: Decision Trees, k-Nearest Neighbors (k-NN, k=3), and Naive Bayes. Comprehensive student data encompassing behaviors, adaptivity levels, and demographic information was analyzed to predict adaptability levels. The Decision Tree algorithm provided a foundational understanding but exhibited limitations in predicting higher adaptability, likely due to overfitting. The k-NN algorithm surpassed others, achieving the highest overall accuracy of 74.80% and demonstrating particular strengths in identifying moderate adaptability levels. This success can be attributed to its ability to recognize subtle similarities among data points, a crucial feature for analyzing nuanced hybrid learning experiences. However, k-NN faced challenges with imbalanced data, as evidenced by a lower recall for high adaptability levels. The Naive Bayes algorithm, despite its lower overall performance, offered valuable insights into the role of feature interdependencies. The study concludes that k-N’s localized pattern recognition provides a more accurate reflection of student adaptability in hybrid learning contexts, highlighting the importance of contextual and relational data analysis.