The increases in releasing high-dimensional radio galaxy images necessitate addressing automatic methods for compacting the curse of dimensionality. Hence, in this paper, the modularized filter-wrapper feature selection scheme (MFWFS) is exploited to select the most discriminative features (MDFs) of the galaxy images. First, we employed the Fanaroff–Riley (FR) radio galaxy images in the data-gathering phase. Next, we applied the MFWFS scheme to 528 moments of radio images for selecting MDFs in dual-phases. The preliminary optimal features (POFs) are selected in the filter phase concerning the triple information theory criteria. In the wrapper phase, the obtained POFs are fed to the twin support vector machine (TWSVM) classifier to extract MDFs. Finally, the quad-final MDFs are introduced to the experimental comparison strategy through the cross-validation technique. Our results show that the best-laid MDFs in a subset of 96 features have accuracy close to 80% with a dimension reduction rate of 5.5.
