A Métrica CFIS

Uma Nova Perspectiva na Análise da Importância das Features em Modelos de Machine Learning

Abstract

The growing demand for machine learning models that deliver quick and accurate responses drives the development of new techniques to enhance model performance and interpretability. In this context, this paper proposes the CFIS (Combined Feature Importance Score), an innovative metric and approach to evaluate feature importance in machine learning problems. CFIS combines different methods, such as permutation importance, regression model coefficients, and correlation with the target variable, to provide a comprehensive view of feature relevance. By integrating these metrics, CFIS aims to overcome the limitations of individual approaches, offering a more robust and detailed analysis of how each feature contributes to model performance. The application of CFIS can benefit various fields, enabling machine learning models to be more transparent and effective in their predictions.