Brain Tumor Classification Using Optimal Features and Ensemble Learning Algorithms

The use of modern technology for better diagnosis of life-threatening diseases like brain cancer is becoming increasingly vital for the medical field. The manual verification of brain images lacks accuracy due to human errors. The Magnetic Resonance Imaging (MRI) dataset contains information on the tumor's existence, and different types of brain tumors like glioma, meningioma, and pituitary. Recent developments in the fields of Machine Learning (ML) can substantially improve the accuracy of medical scanning image (a.k.a MRI images) processing. This paper focuses on utilizing different ML algorithms to examine brain MRI images and identify the presence or absence of tumors. First, the MRI images are enhanced using an adaptive median filter. Then, the MRI dataset is pre-processed using canny edge and Gabor filter to extract features. After that, the dimension of the feature is decreased by applying Principal Component Analysis (PCA) method which decreased the dimension drastically. Several ML algorithms are then applied, namely, logistic regression, Support Vector Machine (SVM) with linear, polynomial, and RBF kernels, ensemble methods Random Forest, Gradient Boosting, and Extreme Gradient Boosting (XGBoost). The experimental results showed that (a) the performance of the ensemble method XGBoost is better than other ML algorithms such as logistic regression, SVM with linear, polynomial, and RBF kernels, Random Forest, Gradient Boosting in terms of accuracy, precision, recall, and F1-score; (b) the XGBoost consumes more training time due to the more number of trees, depth, and iterations, and (c) the experimental results proved that feature reduction using PCA decreased the training time in ML algorithms.