Threshold benchmarking for feature ranking techniques

In prediction modeling, the choice of features chosen from original feature set is crucial for accuracy and model interpretability. Feature ranking techniques rank by its importance but there no consensus on number to be cut-off. Thus, it becomes important identify a threshold value or range, so as remove redundant features. this work, an empirical study conducted identification benchmark algorithms. Experiments are Apache Click dataset with six popularly used ranker machine learning techniques, deduce relationship between total input (N) range. The area under curve analysis shows that ≃ 33-50% necessary sufficient yield reasonable performance measure, variance 2%, in defect models. Further, we also find log2(N) represents lower limit