Machine Learning Based Brain Tumor Classification Using Texture and Statistical Features

Identifying brain tumors possesses a challenging problem in medicine. Researchers typically use a range of traditional diagnostic techniques to increase diagnosis accuracy. However, these procedures can be costly and tedious. The study investigates how well machine learning methods such as SVM (Support Vector Machine), Decision Tree, Naïve Bayes, KNN (K-Nearest Neighbor), LDA (Linear Discriminant Analysis), Ensemble, and ANN (Artificial Neural Network) work for brain tumor diagnosis. Using the GLCM (gray-level co-occurrence matrix) technique, the evaluation function will be converted into numerical values to select appropriate attributes. The study utilizes texture, statistical, and combined feature extraction, as well as raw data without feature extraction, to determine the most accurate machine learning algorithm for identifying and categorizing brain tumors. Upon evaluation of performance, it is observed that the LDA outperformed the existing classifiers. Each algorithm's training time is also investigated. With 200 brain MR images for the study, utilizing the train-test split method, real set of data had training and testing subsets division in a 100:50 ratio. Data were split into both the training and test sets, with 67% allocated for training and test set of 33% for the first six mentioned algorithms. The labeled dataset for the ANN approach is split into three proportions: 70% for training, 15% for validation and 15% for testing.