ARCHIVES

Understanding the true cause-and-effect relationships of factors involved in an academic setting is crucial for the development of effective strategies to enhance the performance of students. Mostly used approaches based on correlation analysis and machine learning models are designed to capture the statistical associations and make predictions, which are not sufficient to support robust decisions about interventions. This study proposes an automated framework which is the integration of three main areas, namely correlation analysis, causal discovery and causal inference, in order to analyze the true causal factors responsible for the performance of students using “student performance dataset” from UCI Machine Learning Repository. As a first step, the correlation based analysis is performed to identify linear association among variables. To show that correlation alone is not enough, causal discovery algorithms are employed to uncover the underlying causal relationship in dataset. Different algorithms such as PC, GES, LiNGAM are used, with LiNGAM based discovery being selected as it produces a fully directed acyclic graph (DAG). As a final step, the discovered causal graph is then used to model the causal inference system to estimate the causal effect of different variables on the performance of students using backdoor adjustment and linear regression algorithm. The results demonstrate some factors like previous performance grades (G1 and G2), study time, paid classes, internet access exhibit a strong causal effect on the final grade of students. Some factors like previous failures, parental status, excessive going out lowers the students' grades. Some features have minimal or negligible impacts on the exam performance. This indicates that intervening on these factors in the academic domain could enhance the performance of students. Compared to correlation based methods, the proposed causation analysis provides directional and interpretable causal relationships, thereby making it easier to understand the underlying causal mechanism and supporting better decision-making.