Imbalanced Multiclass Data Classification Using Combined Data Sampling and Deep Learning Method

Abstract

Multiclass Classification for finding pattern refers to classifying each data to part of the classes or labels that are generally more than two. The foremost challenge in classifying is with imbalanced data that have large portion data known to be majority class, and small portion known as minority class that leads to poor understanding of samples and less accurate results. The existing works discussed Random Upsampling, Random Downsampling, SMOTE methods individually with FeedForward Neural Network and found Random Oversampling gave better results .However, it generates more duplicate data and has less accuracy. Hence , this research work put forward Combined Random Over-Under Sampling approach that was preprocessed prior with Replacing Missing value with mean, Feature selection, Noise Filtering. Meanwhile this work extends the existing FeedForward Neural Network to Deep Learning . The proposed work is implemented in Rapidminer tool, assessed with appropriate evaluation measures for training and testing data individually.