University dropout: Prevention patterns through the application of educational data mining

Abstract

Recently, the use of educational data mining techniques has gained great relevance when applied to performance prediction, creation of predictive retention models, behaviour profiles and school failure, amongst others. For the present paper we applied an attribute selection algorithm to identify the most important factors influencing drop out decision. Decision trees were used to define patterns that can alert an imminent dropout. A tool was adapted and administered online to 300 students from public HEIs, and 200 students from private HEIs currently enrolled on a higher education program. By means of the attribute selection algorithm, 27 relevant factors were found. Within the three main factors, the lack of counselling, an adequate student environment and academic follow-up were recognized, whilst, 7 patterns were found through the decision tree. These included factors such as: student environment, insufficient financial support, experience of an uncomfortable situation and place of career choice, amongst others. Finally, it has been seen that school drop-out does not depend on a single factor but is multifactorial. It is imperative to expand the sample to include other cities. This will enable various algorithms to be applied, providing greater information and leading to the establishment of accurate mechanisms for reducing university drop-out rates, according to the characteristics of the student population in each region.