For decades, predicting the presence of gliomas in MRI images and detecting brain tumours at an early stage have been difficult tasks. As the conventional methods are time-consuming and have the problem of low accuracy, there is a need for an automatic, non-invasive diagnosis system for Giloma brain tumours using machine learning techniques. At first, the input MRI images, collected from the hospital, are enhanced using CLAHE, and the skull removal is done. A collection of features, including a modified local binary pattern, a histogram of Gaussian features, intensity features, and a gray-level co-occurrence matrix, are extracted, and the optimised features are selected using sparse PCA. Finally, radial kernel support vector machines are adopted for the accurate classification of normal and Giloma images after being trained with several images. The experimental results of the proposed method depict that the accuracy, sensitivity, and specificity reach 98.5%, 99.3%, and 97.62%, respectively, which is comparably high compared to many existing automatic systems.