PREDICTION OF ACADEMIC FAILURE IN HIGHER EDUCATION USING DATA MINING TECHNIQUES: THE CASE OF THE INSTITUTO SUPERIOR POLITÉCNICO DO MOXICO

Abstract

School failure in higher education stems from multiple variables linked to the student’s profile and the learning environment. This paper analyses data from a study conducted at the Instituto Superior Politécnico do Moxico, focusing on first-year Computer Science students enrolled in the Programming Logic module, with the aim of evaluating student performance through data mining algorithms to identify predictive patterns of failure. The research adopted a qualitative, descriptive, and action-research approach, involving 20 students. Data gathered through observation, document analysis, questionnaires, and interviews were processed using WEKA. Five classification algorithms (J48, REPTree, NaiveBayes, SMO, and Multilayer Perceptron) were compared via 10-fold cross-validation. NaiveBayes stood out, achieving 90% accuracy by correctly classifying 18 of the 20 instances. The pre- and post-intervention comparison, based on 13 interactions, revealed a systematic reduction in the predicted failure rate: initial values ranging from 50% to 96% fell to between 1% and 68%. The mean decreased from 78.77% to 18.15%, representing a variation of 60.62 percentage points. These findings underscore the utility of data mining in mapping academic risks and monitoring predictive trends, whilst noting that the study design does not permit the inference of direct causality between the implemented actions and the observed reduction.