Comparison of text vectorization techniques for machine learning applied to binary classification

Lopez-Robles J.E.; Morgan-Benita J.A.; Luna-García H.; Reveles-Gomez L.C.; Celaya-Padilla J.M.; Galvan-Tejada C.E.; Flores-Chaires L.A.; Acra-Despradel C.; Villalba-Condori K.O. · PeerJ Computer Science (2026) · DOI: 10.7717/peerj-cs.3889

Resumen

Text vectorization is indispensable for natural language processing (NLP). It allows us to transform text—from comments, reviews, documents, and even complete books—into formats or values that we can use as input in machine learning (ML) models. In this article, we present a comparative analysis between some of these methods used for text vectorization. The general purpose is to identify which of these methods offers a superior performance when applied together with some classification models on binary classification datasets. Each vectorization method, in combination with the models used, is evaluated by standard metrics in machine learning, such as Accuracy, Precision, Recall, F1-score and area under the curve (Receiver Operating Characteristic (ROC)) for a robust evaluation, demonstrating that the newest and most complex vectorization methods are not necessarily better for all binary classification tasks. Copyright 2026 Lopez-Robles et al. Creative Commons CC-BY 4.0. http://www.creativecommons.org/licenses/by/4.0/

Instituciones: Universidad Católica de Santa María