Predictive Models for Early Detection of Engineering Students at Risk of a Course Failure

© 2022 IEEE.In this study, the creation of predictive models for the detection of students at risk of failure is presented. A sample of 618 student profiles , 19% of which failed a given course, were used to detect students at risk of failing. The student profiles were determined from the constructs of Self-Regulation Learning and Affective Strategies (SRLAS) and Multiple Intelligences (MI). The first part of this work, includes an Exploratory Factor Analysis of the data. The predictive phase of the study uses nine Machine Learning classification techniques (KNN, SVM, LDA, QDA, Decision Trees, Random Forest, ADA_Boosting, XGBoosting, and Bayes) to classify students that passed or failed a given course. The Log-Math and the Anxiety dimensions turned out to be relevant variables regarding the success of the student. Decision Trees and Random forests provided the best predicting power. Our results are encouraging in the sense that the methodology followed proves successful at identifying the main factors related to student failure. This may help instructors to timely identify students at risk of failure and their possible causes, to implement appropriate strategies to mitigate this undesirable outcome.

Predictive Models for Early Detection of Engineering Students at Risk of a Course Failure Academic Article in Scopus