Modelo Predictivo para la Deserción Estudiantil

Abstract

Abstract: This investigation presents the development of a predictive method to face the student dropout in highest education using for this effect, advanced techniques of Artificial Intelligence mainly related to Machine Learning. It follows the CRISP-DM methodology, it analyzes a group of administrative and academic data applying characteristics engineering and SMOTE technique to balance the classes. The comparative analysis reveals the Random Forest optimized as the more effective achieving an accuracy of 84.09% and AUC of 0.92. Besides it is included a classification analysis to discover trends with students which eventually can dropout their studies. The results obtained from the study are benchmarked with a study done in Iceland. With this comparation it can be concluded academic performance and finance conditions are the main indicators in common. An important contribution is the identification of risk profiles using no supervised learning with the algorithm K-Means in order to do early interventions to prevent students to dropout their careers. This vision not only improve retention inside Institutions but create a technical standard to be used to strengthen educative policies inside the institutions based on the results here obtained.