Automatic Retraining Approach of a Machine Learning Model for Prediction of the COVID-19 Infection Rate

Gutierrez E.P.; Vilca D.C. · Computacion y Sistemas (2026) · DOI: 10.13053/cys-30-2-5352

Resumen

With the arrival of the COVID-19 infection worldwide, the economic and health stability of various countries was affected. Despite numerous proposals for predicting COVID-19 infection, the problem of the short-lived accuracy of these models persists. This article presents an approach for the automatic retraining of a COVID-19 infection rate prediction model for any country in the world. The aim is to provide a tool for protocol and strategic planning to counter this infection, with accuracy that endures over time. The model was developed using a recurrent neural network (LSTM) and a constantly updated data source, essential for the application of the approach. The models produced by this approach maintained an average coefficient of determination of 0.989, 0.986, and 0.996 in three different countries— Peru, Brazil, and Chile, respectively—during a month of weekly retraining. Compared to other studies, the models produced have the advantage of being trained weekly with updated data, maintaining long-term accuracy. Furthermore, the effectiveness of an automated machine learning model retraining process was verified.

Instituciones: Universidad Nacional Mayor de San Marcos