An Evaluation of Feature Selection Methods on Multi-Class Imbalance and High Dimensionality Shape-Based Leaf Image Features

Abstract

Multi-class imbalance shape-based leaf image features requires feature subset that appropriately represent the leaf shape. Multi-class imbalance data is a type of data classification problem in which some data classes is highly underrepresented compared to others. This occurs when at least one data class is represented by just a few numbers of training samples known as the minority class compared to other classes that make up the majority class. To address this issue in shapebased leaf image feature extraction, this paper discusses the evaluation of several methods available in Weka and a wrapperbased genetic algorithm feature selection.