KDE-Based Ensemble Learning for Imbalanced Data

Imbalanced class distribution affects many applications in machine learning, including medical diagnostics, text classification, intrusion detection and many others. In this paper, we propose a novel ensemble classification method designed to deal with imbalanced data. The proposed method trains each tree in the ensemble using uniquely generated synthetically balanced data. The data balancing is carried out via kernel density estimation, which offers a natural and effective approach to generating new sample points. We show that the proposed method results in a lower variance of the model estimator. The proposed method is tested against benchmark classifiers on a range of simulated and real-life data. The results of experiments show that the proposed classifier significantly outperforms the benchmark methods.