Impacto de las estrategias de preprocesamiento de datos en la precisión de algoritmos de regresión lineal

Abstract

Data preprocessing is a crucial stage in the performance of linear regression models, where data quality including the handling of missing values, outlier detection, normalization, and variable encoding directly influences the predictive capability of machine learning algorithms. This systematic review, conducted under the PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) methodology, analyzes 40 scientific articles published between 2021 and 2026 in well-recognized databases such as Scopus, Web of Science, Springer Nature, MDPI, IEEE, and ScienceDirect. Within this context, linear regression models are highlighted as a fundamental tool in artificial intelligence and data analysis, where the quality of preprocessing directly affects the accuracy of the results obtained. During the development of the study, key techniques were examined, including data cleaning, missing value imputation, normalization or standardization, as well as the encoding of categorical variables. Regarding practical applications and future research directions, the findings support their implementation in predictive models related to sales, price estimation, and academic indicators. They also encourage extending the analysis to non-linear models and higher-dimensional datasets. Ultimately, proper data preprocessing significantly improves the predictive accuracy of linear regression models, helping to reduce issues such as overfitting and underfitting.