Mutual Information-Based Feature Selection and Ensemble Learning for Classification

Feature selection approaches aim to maximize relevance and minimize redundancy to the target by selecting a small subset of features in classification. This paper proposes a feature selection method based on mutual information (MI). We select a feature subset with minimal redundancy maximal relevance criteria. Multiple kernel learning (MKL) and ensemble learning (EL) have been applied in hyperspectral image classification. Our method applies Adaptive Boosting (AdaBoost) approach to learning multiple kernel-based classifier for multi-class classification problem. Classification experiments with a challenging Hyperspectral imaging (HSI) task demonstrate that our approach outperforms current state-of-the-art HSI classification methods.