Nowadays, improving classification performance of the hyperspectral image is an important task due to their wide range of applications in the field of land cover identification, precision agriculture, and biomedical Imaging. Nevertheless, several constraints often diminish the performance such as curse of dimensionality, limited labelled samples high inter-class similarity and intra-class variability. In this paper, we propose an efficient Radial Basis Function (RBF) Kernel based Extreme Learning Machine (KELM) to ensure fast and reliable classification model. We explore local spatial features by utilizing Local Binary Patterns (LBP) and Gray-Level Co-occurrence Matrix (GLCM). To reduce redundant spectral bands, the JAYA optimization algorithm is employed with a composite fitness function that combines classification accuracy and the Jeffries–Matusita (JM) distance ensuring optimal class separability bands. In addition, Principal Component Analysis (PCA) is applied to the selected bands to extract first three principal spectral components while preserving both global and local texture information to derive spatial features. Finally, the selected spectral and spatial features are concatenated and normalized to produce a compact feature set as input to KELM classifier. Expensive experiments conducted on three benchmark HSI datasets including Indian Pines, Pavia University, and Salinas to validate the efficiency of the proposed approach. The experimental results demonstrate competitive performance over SOTA techniques in terms of overall accuracy, average accuracy, and class-specific performance.