Comparativa de modelos de clasificación para inferir la probabilidad de deserción estudiantil en la Facultad de Ciencias Químicas de la Universidad de Cuenca

Abstract

This degree work shows an application of comparative classification models, through specific variables, to determine the university dropout of students from the Faculty of Chemical Sciences of the University of Cuenca. In this context, through data mining, two classification models were applied: K- nearest neighbors (knn) and logistic regression to classify first-year students into two populations: dropout or permanence. The data was obtained from the socio-economic record of the students from 2014 to 2018, in addition, the population groups corresponding to those who dropped out in the first year and those who continued with their studies were identified. Based on this, it was possible to interrelate the variables to group them through principal component analysis (PCA). The data were separated for training and validation of the models. The systems were modeled in RapidMiner generating a confusion matrix, which allowed determining that the knn model presents a better current of 73.30% compared to 54.67% of the Logistic Regression model. Additionally, it was concluded that the most relevant variables are those that make up the main component 1: total income, total expenses, monthly rent payment, type of high school, cumulative valuation of vehicles. Through the confusion matrix, the models (knn and rl) were evaluated, selecting the knn model as the best option.