Predictive analysis of dropouts in Information Technology higher education

Abstract

The study primarily aims to identify the key attributes that contribute to student dropouts in ICT higher education, a significant issue in educational data mining. It seeks to explore the distinct factors influencing dropout rates that have been underexplored in existing literature. To achieve this, an experimental approach was employed to create a comprehensive database for extracting relevant insights. Data was collected from five batches of students enrolled in the information technology course at a government tertiary education institute in Sri Lanka. The collected data underwent pre-processing, which involved imputing missing values, transforming data formats, and selecting relevant variables. Feature selection was carried out using correlation-based feature selection (CFS) to pinpoint subsets of attributes closely linked to dropout outcomes. Mostly used classification algorithms were evaluated based on their performance using confusion matrix metrics. After evaluating mostly used classification algorithms, this study trained multiple classification models, including Decision Tree, K-Nearest Neighbor, Naïve Bayes, and Rule-Based approaches, to identify strong relationships between dropout factors and dropout status, achieving over 86.2% accuracy. The J48 Decision Tree emerged as the most accurate algorithm for the dataset and was used to build a predictive model for student profiles. The performance of the model was validated using a new dataset sourced from institutional records. The dropout prediction application was implemented using the Java WEKA API and achieved 92.6% accuracy in predicting student dropouts in ICT higher education. By uncovering strong relationships between dropout factors and dropout status, the study highlights key influences, with the most significant factors being perceived course quality, prior academic qualifications, prior ICT experience, 0/L results, and English proficiency score. This model can be utilized to predictively analyze student dropouts in higher education, allowing early identification of at-risk students and facilitating targeted intervention strategies.