Preventing Overfitting and Underfitting in Machine Learning Model Development: A Practical Analysis

Abstract

The main objective of this document is to assist future researchers in identifying fit problems in the development of predictive models used in Machine Learning. To this end, theoretical aspects were addressed, and most importantly, a practical case study was developed using synthetic data. Three predictive models were created from this data: an underfitted model, an overfitted model, and an ideal model. This allowed for the identification of the characteristics of each type of fit. Finally, a simple and practical guide was created outlining the steps to follow to obtain a model with an adequate fit, that is, one with good generalization capabilities.