Improving Class Imbalance Detection And Classification Performance: A New Potential of Combination Resample and Random Forest

Abstract

Data mining is a knowledge discovery of the data that extracts and discovers patterns and relationships to predict outcomes. Class imbalance is one of the obstacles that can drive misclassification. The class imbalance affected the result of classification machine learning. The classification technique can divide the data into the given class target. This research focuses on four pre-processing methods: SMOTE, Spread Subsample, Class Balancer, and Resample. These methods can help to clean the data before undergoing the classification techniques. Resample shows the best result for solving the imbalance problem with 41.321 for Mean and Standard Deviation, 64.101. Besides, this research involves six classification techniques: Naïve Bayes, BayesNet, Random Forest, Random Tree, Logistics, and Multilayer Perceptron. Indeed, the combination of Resample and Random Forest has the best result of Precision, 0.941, and ROC Area is 0.983.