School Dropout Prediction Model Based on Undergraduate Course Self-Assessment Data

Abstract

School dropout is a daily challenge for educational institutions. In the specific case of higher education, high dropout rates lead to financial losses and a shortage of professionals in the market. This research aimed to develop and evaluate a predictive model to identify students prone to dropout, using data from a semester-based self-assessment model of undergraduate courses at Federal University of Paraíba (UFPB). The study analyzed the relationship between school dropout and institutional self-assessment by utilizing educational data mining and the CRISP-DM methodology, followed by exploratory analysis and data preparation for classification. Various modeling techniques, such as Decision Trees, Random Forest, and Support Vector Machines, were applied, with the models evaluated using performance metrics, revealing an accuracy of 87.97%, precision of 91.72%, recall of 91.67%, and F-measure of 91.57% in identifying students with a high probability of dropout. Approximately 59% of active students at UFPB, admitted from 2017 onwards, showed a high probability of abandoning their courses in the proposed predictive model tests. This information can support institutional decisions and the implementation of effective policies and actions against dropouts, aiming to improve academic outcomes. The study contributes to advancements in predicting school dropout, providing valuable insights for decision-making and preventive strategies in UFPB and other higher education institutions.