Fig. 24

Power-law analysis of the dataset using CCDF on a log–log scale, showing that the empirical distribution follows a heavy-tailed pattern where a few high-impact features dominate predictive influence, while most features contribute modestly, consistent with power-law behavior.