Logistic Regression and Random Forest for Effective Imbalanced Classification

Abstract

Nowadays, the application of data mining and machine learning techniques continues to be common in many fields. There are many imbalanced datasets with much less significant samples than unimportance ones in real-life because it is hard to collect representative positive examples. Under these circumstances, the conventional aim of reducing overall classification accuracy and most of the standard machine learning methods may not be suitable for the imbalanced problem. In this work, we compare the performance of random forest and logistic regression on the prediction of an imbalanced dataset. We propose several ways to enhance two models based on cost-sensitive learning to provide more accurate predictions when dealing with imbalanced datasets.