Applying survival analysis and Explainable Artificial Intelligence to understand academic success in software engineering students

Abstract

This paper presents a characterization of software engineering students based on their admission profiles and academic trajectories. Using a larger dataset from the 2014–2019 cohorts, the study combines statistical methods, supervised learning techniques, Explainable Artificial Intelligence (XAI), and Survival Analysis (SA) to understand and predict student outcomes. The results confirm a strong relationship between admission variables, such as age, CENEVAL score, GPA, credit progress, and intended highest educational level, and academic success or dropout risk. Multinomial Logistic Regression, Decision Trees, and Cox Proportional Hazards models were employed to classify students into three academic trajectory (AT) types: timely graduation (Type A), delayed graduation (Type B), and dropout (Type C). Multinomial Regression achieved the highest accuracy at 90%, outperforming the other models by approximately 10%. Notably, while a high CENEVAL score often indicated successful graduation, exceptions were observed, emphasizing the need for more nuanced models. Survival Analysis provided a dynamic view of academic trajectories over time, revealing that student performance can shift due to various internal and external factors. These findings offer valuable insights for educational authorities, enabling targeted interventions to reduce dropout rates and improve program retention. Future work will explore additional XAI techniques, expand the student sample size, incorporate new variables like student motivation, and perform comparative studies across different Mexican regions to evaluate geographic influences on software engineering student profiles.

Ähnliche Arbeiten