Análise de Interações em dados dietéticos: uma abordagem via LASSO hierárquico

Abstract

It is well known that the combination of dietary factors, lifestyle, clinical conditions, and sociodemographic characteristics plays an important role in human health. In an effort to understand these relationships, traditional linear models are widely used to identify and quantify interactions between variables. However, as the number of variables and potential interactions increases, these models face limitations, such as increased dimensionality, risk of overfitting, and the selection of interaction effects that do not satisfy the hierarchical relationship between main effects and interaction effects. In this context, regularized regression techniques, such as LASSO, have emerged as promising alternatives because they enable automatic variable selection and produce simpler and more interpretable models. Hierarchical LASSO, in particular, represents an important advancement, as it imposes structural constraints that ensure consistency in the inclusion of interaction effects, leading to more reliable results. Therefore, this study aimed to investigate the relationships between dietary and lifestyle variables associated with insulin resistance by evaluating the effectiveness of Hierarchical LASSO in identifying factors associated with the HOMA-IR index. To this end, both a theoretical and an applied analysis were conducted using a real-world dataset and the R statistical software. The results showed that Hierarchical LASSO identified relevant main effects and interaction effects among dietary, anthropometric, and clinical variables, achieving better predictive performance than Classical LASSO for the analyzed data and demonstrating its potential as a promising tool for improving the understanding of insulin resistance.