A Smart Approach for Imbalanced Data Using Data Sampling and Binary Harris Hawks Optimisation

Abstract

Data preprocessing for imbalanced datasets still becomes an essential stage in machine learning studies whenever the classes are uneven in the training data. The class imbalance may vary to extents but when the imbalance is extreme it is challenging to model and may require advanced methods. Thereby, if the data set is unbalanced then it would predict the most frequent class goodbye all with zero percent accuracy of the least frequent class which is generally the goal of developing the model in the first place. Class balance is the main reason why oversampling is used as a popular technique. Furthermore, the majority of feature selection algorithms make an inaccurate feature estimation by estimating a feature's significance for a class without taking the dataset's imbalance into account. In this learning, we suggest a proficient model namely ASBHHO for features selection using binary Harris Hawk Optimization algorithm accompanied with Synthetic Minority Oversampling Technique. Analysis of the findings of our experimental learning verifies the efficiency of the ASBHHO algorithm in practice. To support the above statements about the precision and effectiveness of the method proposed, its performance is related to other advanced methods.