Combining SMOTE and Intuitionistic Fuzzy rough assisted model for data reduction to handle imbalanced datasets

Abstract

Due to continuous bombardment from different sources such as social media like Twitter, Facebook, Instagram etc. and day by day activities, the size of datasets is gradually increasing. Data is enlarging both in number of instances and features. Processing these datasets using traditional algorithms and techniques require more resources and exponential time. Moreover, these datasets usually consisted of various issues such as class imbalance, availability of irrelevant and/or redundant features, noise, problematic instances and different types of uncertainties that may lead to degrade the overall performances of the learning algorithms. Our aim is to keep the maximum number of informative features and instances that could help for distinguishing the different categories without conflict. This paper presents a novel approach by combining synthetic minority oversampling technique (SMOTE), Intuitionistic Fuzzy (IF) set theory, rough set theory (RST), different class ratio, and sample pair selection concepts to handle all the above mentioned issues. Firstly, we incorporate SMOTE with a new IF rough instance selection (IS) for the cleaning of problematic minority and majority class instances. Secondly, a novel IF rough feature selection (FS) by using different class ratio based on the concept of sample pair selection is presented to eradicate irrelevant and redundant features along with noise. Thirdly, a comprehensive experimental study is given to justify the effectiveness of the proposed techniques. Finally, we present a framework to enhance the prediction of blood–brain barrier penetrating peptides (bbbpp) to expedite drug delivery into the brain.