Comparing Two Categorical Gini Correlations with Applications to Classification Problems
A new statistical framework is proposed for comparing predictor importance in classification tasks with categorical outcomes. The method utilizes the categorical Gini correlation (CGC) to assess the dependence between numerical predictors and categorical results. This approach can handle predictors of varying dimensions and dependencies, and its statistical properties are analyzed, including asymptotic normality and consistency, with a nonparametric bootstrap procedure also offered for inference. AI
IMPACT Introduces a novel statistical method for evaluating predictor importance in classification problems, potentially improving model interpretability and feature selection in AI applications.