For surgical robotics to guide robotic arms during surgeries, precise tumour detection and segmentation are crucial. The study’s feature extraction techniques and machine learning models can be included into image-guided surgical robots to improve their accuracy when performing brain tumour surgeries, especially when it comes to distinguishing between normal and aberrant tissues. In medical imaging, identifying post-operative recurrent brain tumours is crucial, and accurate diagnosis frequently necessitates the use of complex techniques. The EPISURG dataset, used in this work, has 430 people, 381 samples labelled “no tumour” and 49 samples labelled “tumour”. Using an 80–20 train-test split, about 344 training samples and 86 test samples were obtained. Using Grey Level Co-occurrence Matrix (GLCM) and Local Binary Patterns (LBP) for feature extraction, we assessed the performance of several classifiers. Among these classifiers were Support Vector Machines (SVM) and Random Forests (RF), both with and without the Radial Basis Function (RBF) Kernel. The findings showed that the SVM with RBF Kernel using LBP features had a test accuracy of 0.70 while the GLCM features had a test accuracy of 0.82. We used Principal Component Analysis (PCA) on the combined data to reduce dimensionality, and the SVM with RBF Kernel produced the best test accuracy of 0.86. However, the fully connected layer model encountered overfitting errors. This work highlights the role that feature selection and dimensionality reduction techniques play in enhancing model performance and demonstrates how well-suited handcrafted feature extraction techniques are for improving the accuracy of identifying recurrent brain tumours.