Soft computing techniques to determine the factors that influence student dropout

Abstract

The study conducted at the Technical State University of Quevedo delved into student dropout by examining academic, socioeconomic, and personal data through Soft Computing techniques, utilizing 225 variables and 160,000 records. Algorithms such as Random Forest, Gradient Boosting, and SVM were compared, with Random Forest standing out for its precision in predicting critical factors influencing dropout. Among the determining factors are the cut-off grade 2, the head of household's occupation, and the student's age, identified as the most significant for student performance. The findings of the study highlight the complexity of student dropout, emphasizing that students at greater risk of failing tend to have low mid-term grades, come from households with unemployed heads or in unskilled jobs, and are in the age range of 40-45 years. On the other hand, student profiles with higher chances of passing show opposite patterns in these variables. These results underline the need for adopting comprehensive and personalized approaches to address dropout, considering both academic factors and socioeconomic and personal contexts.

Keywords:

Prediction, University Education, Artificial Neural Networks, Predictive Models.