Minkowski distance measure in fuzzy PROMETHEE for ensemble feature selection

Abstract

A data preprocessing step is necessary to modulate effective and efficient data that can be utilized in numerous data mining and machine learning problems, especially in high-dimensional datasets. Most of the feature selection methods are unstable as different subsets provide various subsets of features giving different classification accuracy. Through ensemble feature selection, higher accuracy can be achieved and it has been verified theoretically and experimentally that diversity among base classifiers further improves its accuracy. In this paper, we model the ensemble feature selection methodology as a decision-making technique. We consider Minkowski distance for fuzzy preference ranking organization method for enrichment evaluation (PROMETHEE) decision making problem as the ensemble feature selection problem. Due to its main advantages in minimizing scalability between criteria and its ease of use for mathematical computations that outrank alternatives, PROMETHEE is an effective tool. To establish the superiority of the proposed methodology, a comparison between existing methodologies has been carried out through various performance metrics.