Analysis of Class Imbalanced Brain Tumor Using Machine Learning Techniques

Abstract

It is evident that healthcare has become a critical global priority, and the intelligent utilization of clinical datasets is essential for establishing an effective and efficient healthcare system capable of monitoring and managing people's health. However, the issue of class imbalance in real-world datasets, including clinical datasets, poses significant challenges to the training of classifiers and can result in reduced accuracy, precision, recall, and increased misclassifications. In our comprehensive literature review, we've examined the performance of five well-known classifiers—Logistic Regression, Decision Tree, Gaussian Naive Bayes, Random Forest, and Extra Tree classifiers—over imbalanced brain tumor datasets. We have also evaluated the effectiveness of four different class balancing techniques—SMOTE, ADASYN, ENN, and SMOTE-ENN—in addressing the challenges posed by imbalanced class distributions. The results of our study indicate that the SMOTE-ENN balancing approach has demonstrated superior performance compared to the other three data balancing strategies when used with all five classifiers. Additionally, although the other three balancing strategies, namely SMOTE, ADASYN, and ENN, performed relatively well, they slightly underperformed in comparison to the SMOTE-ENN approach. The identification of the SMOTE-ENN approach as the most effective strategy for handling imbalanced datasets is significant, as it highlights the importance of combining over-sampling and under-sampling techniques to achieve a more balanced and representative dataset for training classifiers. By effectively addressing the issue of class imbalance, the SMOTE-ENN approach allows for the development of more robust and accurate predictive models, thus improving the overall performance of the classifiers on imbalanced brain tumor datasets. Our study contributes valuable insights into the selection of appropriate data balancing strategies and classifier choices when dealing with imbalanced datasets in the healthcare domain. By providing a comprehensive overview of the empirical performance of different classifiers and balancing techniques, we have laid the foundation for implementing more effective and reliable supervised machine learning algorithms in the context of clinical data analysis. The recommendations we offer for dealing with class imbalanced datasets further enhance the practical applicability of our research findings.