Detection of Brain Tumor using Boosting Algorithms based on Feature Selection

Brain tumors are one of the most common causes of death. An early and correct identification of brain tumors is critical for effective therapy. Using artificial intelligence-based software programs instead of traditional methods can provide more accurate results in brain tumor detection. Especially recently, there have been many studies in the detection of diseases based on the processing of medical images. In this study, a novel hybrid algorithm was proposed based on three different feature selection algorithms (univariate feature ranking for classification using chi-square tests (f-chi2), rank the importance of features using ReliefF algorithm (f-Relief), rank features for classification using minimum redundancy maximum relevance algorithm (f-mRMR), and the classic and ensemble learning, respectively based on support vector machine (SVM) with different kernel structures and ensemble learning (EL) with boosting methods, were performed to detect the brain tumor using magnetic resonance imaging (MRI) features. K-fold is used to prevent overfitting. Analysis results show that a 100% accuracy score was achieved in the ensemble-based classifier in the detection of brain tumors with the proposed hybrid method. As a novelty for detecting the tumors, statistics-based feature selection methods are proposed, to help reduce the size and thus reduce complexity in complex network problems. The proposed method suggests a feature selection algorithm that can help reduce the data size in future studies.