A comparative analysis of machine learning approaches to predict brain tumors using MRI

A brain tumor is an excessive development of dangerous cells in the brain and is one cause of cancer deaths worldwide. Early detection is important for the effective treatment of brain tumors. To do this, magnetic resonance imaging is used as a safe and non-invasive method. In recent years, machine learning (ML) has attracted more attention due to its achievements in the field of disease diagnosis and precision medicine. Many studies have proposed using ML to predict the onset of a brain tumor. The main objective of this analysis was to determine whether ML algorithms can predict the onset of brain tumors accurately and effectively. We hypothesized that ML algorithms would provide an accuracy of more than 90%, and an ensemble model would provide better accuracy within 30 minutes of computational time. Among various classification methods, we compared the two most popular ML algorithms, support vector machine (SVM) and random forest (RF), to provide some insight on model selection for brain tumor classification. To test the generalizability of each ML algorithm on independent datasets, we used cross-validation to evaluate model performance. To further improve the model performance, we introduced grid search to find the optimal values of hyperparameters for the classifiers. We observed the trade-off between performance improvement and time consumption of grid search. Finally, we developed an ensemble model which integrated k-fold cross-validation with the SVM and RF classifiers to achieve the highest model accuracy with less time and make the traditional ML methods more robust.