On predicting school dropouts in Egypt: A machine learning approach

Abstract

Compulsory school-dropout is a serious problem affecting not only the education systems, but also the developmental progress of any country as a whole. Identifying the risk of dropping out, and characterizing its main determinants, could help the decision-makers to draw eradicating policies for this persisting problem and reducing its social and economic negativities over time. Based on a substantially imbalanced Egyptian survey dataset, this paper aims to develop a Logistic classifier capable of early predicting students at-risk of dropping out. Training any classifier with an imbalanced dataset, usually weaken its performance especially when it comes to false negative classification. Due to this fact, an extensive comparative analysis is conducted to investigate a variety of resampling techniques. More specifically, based on eight under-sampling techniques and four over-sampling ones, and their mutually exclusive mixed pairs, forty-five resampling experiments on the dataset are conducted to build the best possible Logistic classifier. The main contribution of this paper is to provide an explicit predictive model for school dropouts in Egypt which could be employed for identifying vulnerable students who are continuously feeding this chronic problem. The key factors of vulnerability the suggested classifier identified are student chronic diseases, co-educational, parents' illiteracy, educational performance, and teacher caring. These factors are matching with those found by many of the research previously conducted in similar countries. Accordingly, educational authorities could confidently monitor these factors and tailor suitable actions for early intervention.