Comparative Analysis of Traditional Machine Learning and Deep Learning Techniques for Brain Tumor Identification in MRI Data

Patient outcomes depend on the ability to identify brain tumors from MRI images. Using the BraTS Africa dataset, this study examines a hybrid approach that combines deep learning and classical machine learning for the purpose of segmenting and classifying tumor regions from brain MRI images. Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Convolutional Neural Networks (CNNs) are the five common algorithms that were compared. A preprocessing pipeline emphasized the edges of tumors and enhanced image quality. It featured Otsu and adaptive thresholding, contrast stretching, erosion and dilation morphological operations, Sobel and Laplacian edge detection, greyscale normalization, and histogram equalization. To further enhance feature extraction, advanced methods such as watershed segmentation and Histogram of Oriented Gradients (HOG) were applied. CNNs outperformed Random Forest classifiers, achieving an accuracy of 82.86% and F1-score of 0.81, compared to Random Forest’s accuracy of 77.14% and F1-score of 0.79. The study frames the classification problem as a binary diagnostic task, aiming to algorithmically differentiate between pathological (tumor) and non-pathological (healthy) brain regions in MRI imagery. KNN and Logistic Regression yielded lower accuracies of 51.43% and 60%, respectively, underscoring the enhanced discriminative power of modern AI architectures in neuroimaging diagnostics. The proposed hybrid framework demonstrates the efficacy of integrating advanced preprocessing with model-specific strengths to optimize tumor detection. Personalized cancer treatment plans, radiologists' support, and early diagnosis would all be enhanced by combining these methods with clinical decision-support systems.