In Hyperspectral Image (HSI) classification, various features such as spectral, spatial, texture, shape and statistical features, are exploited to characterize the pixels from different perspectives. Research shows that a careful selection of multiple features can result in better classification performance. In this work, we propose a framework for selection of a most relevant subset of features in an optimal way for HSI classification. When provided with a feature set and evaluation criteria, the framework can rank the features in groups instead of ranking them individually. A group with higher rank be made up of n highest ranked features each for one class. A group of features with mth rank is highly discriminative than the group of features with (m+1)th rank. Therefore, the proposed framework selects the combination of top k groups of features for further processing. A subset of features belongs to each group are supposed to be capable of discriminating subtle classes. Moreover, the proposed framework eliminates the redundant as well as noisy features effectively. Experiments conducted on Pavia University dataset with a support vector machine classifier demonstrate the improved classification performance in comparison to the state of the art feature selection approaches. © 2018 Proceedings - 39th Asian Conference on Remote Sensing: Remote Sensing Enabling Prosperity, ACRS 2018