Predicting Students' Academic Failure Using Machine Learning Algorithms: An Analytical Study Based on Large-Scale Educational Data

The fast adoption of new AI and learning analytics technologies, as well as the widespread usage of digital learning platforms, have allowed the education industry to undergo a fundamental transition. This has enabled the analysis of massive amounts of educational data to support academic decision-making and improve learning outcomes. The use of artificial intelligence and machine learning technologies to create predictive models that can anticipate students' academic challenges early on and enhance the efficacy of educational interventions by empowering educational institutions to proactively assist students who are at risk of academic failure or withdrawal has become more popular as a result of this development. Using extensive real-world educational data from the Open University Learning Analytics Dataset (OULAD), which comprises 32,593 student records, this study attempts to create a prediction model based on machine learning algorithms to forecast students' academic failure. To identify pupils at risk of academic failure in comparison to successful students, an experimental technique was used, which included data processing, behavioral and academic characteristic extraction, and the creation of a binary classification model. In order to distinguish between high-achieving children and those at risk of academic failure, the study used a binary categorization approach. It included a sequence of preprocessing steps and the extraction of academic and behavioral features. Statistical assessment measures were devised to compare five machine learning algorithms (Random Forest, Support Vector Machine, Decision Tree, Logistic Regression, and K-Nearest Neighbors). The results indicated that Random Forest performed better than the others, with an accuracy value of 89.43% and an AUC of 0.9633. This indicates that the model is highly discriminative between academically high-achieving students and those at risk of academic failure. Moreover, the Feature Importance analysis indicated that both continuous assessment performance scores and degree of digital interaction with the learning platform were the most important predictors for academic performance. Our findings further highlight the promise for machine learning technologies to play an important role in educational early warning systems and help develop more adaptive learning environments that serve students' needs.