PROPOSAL OF A TIME SERIES-BASED MODEL FOR THE CHARACTERIZATION AND PREDICTION OF DROPOUT RATES AT THE NATIONAL OPEN AND DISTANCE UNIVERSITY

Abstract

Dropout rates are a key indicator of educational quality, making it imperative for educational institutions to design strategies to reduce them, thereby contributing to improved student retention and the achievement of academic objectives. While dropout research has primarily focused on machine learning methods applied to in-person education datasets, this article introduces a novel approach based on time series models for dro pout rates analysis at the National Open and Distance University (UNAD). Methodologically, an adaptation of the CRISP-DM methodology was undertaken in four phases, namely: F1. Business and data understanding, F2. Data preparation, F3. Model building and evaluation, and F4. Model deployment. In terms of results, an open dataset on UNAD dropout, ob tained from the SPADIES system between 1999 and 2021, was employed. Using Python libraries statsmodels and pandas, an ARIMA model was implemented, displaying optimal error metrics. This ARIMA model was utilized to forecast future dropout rates at UNAD, projecting a future dropout rate fluctuating around 23%. In conclusion, the ARIMA model developed for UNAD stands as an innovative and essential tool in the educational realm, capable of accurately anticipating dropout rates for upcoming semesters. This provides UNAD with a unique advantage in strategic decision-making.