Application of Time Series Models in the Characterization of Dropout at the University of Cartagena, Colombia

Abstract

One of the key factors affecting educational quality in Higher Education Institutions is student dropout. A high dropout rate can indicate student dissatisfaction with the relevance and quality of the education they are receiving. Therefore, a pressing challenge for these institutions is to characterize the dropout rates of academic programs, as well as overall dropout rates, with the aim of identifying trends or potential future variations that could support strategic decision-making to mitigate student dropout. In this regard, it has been generally observed that predictive models have predominantly employed machine learning techniques to predict whether a student will drop out based on social, economic, academic, and demographic variables, without focusing on characterizing percentage variations or future dropout trends. Thus, this article proposes an ARIMA-based time series model as a contribution to the characterization of historical dropout rates at the University of Cartagena, Colombia, from 1998 to 2022, with the goal of forecasting dropout rates for future years. This study was developed through four methodological phases: P1. Creation of training and testing datasets, P2. Identification of the model parameters p, q, and d, P3. Adjustment of potential models, P4. assessment of the ARIMA model and P5. Forecasting future dropout rates. The results showed that the ARIMA model (p=1, d=2, q=0) provided the best fit, enabling predictions to be made up to the second semester of 2028. A key conclusion of the study is that the dropout rate at the University of Cartagena over the next four years is expected to hover around 6%, meaning that for every 100 students entering the university, approximately 6 will drop out.