Enhanced Multi-Class Brain Tumor Classification Model using MRI images: A Wavelet and Deep Learned Features based Approach

Accurate detection of brain tumors from Magnetic Resonance Imaging (MRI) scans is a critical need in modern healthcare, as early diagnosis and precise tumor typing can greatly influence treatment strategies and improve patient prognosis. In this research, we introduce a hybrid classification framework that fuses handcrafted wavelet-based features with deep CNN embeddings extracted from a pre-trained VGG19 network. After an extensive pipeline of image preprocessing, wavelet decomposition, and CNN-based feature extraction, we employ a stacked ensemble classifier composed of XGBoost, Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors, with XGBoost serving as a meta-learner to optimally combine the predictions of all base models. Our experiments on a comprehensive dataset of 1311 MRI images demonstrate a final test accuracy of approximately 92.14%, bolstered by highly positive precision-recall metrics. Detailed discussion of the confusion matrix and classification report reveals that notumor and pituitary cases are most consistently distinguished, whereas glioma and meningioma classes exhibit some degree of overlap, aligning with well-known clinical challenges. Overall, these findings reinforce the viability of combining handcrafted textural features with learned deep representations for accurate and robust neuro-oncological decision support.