Enhanced Brain Tumor Classification Using EfficientNetB0 and Hybrid Machine Learning Approaches

Abstract

The classification of brain tumors has remained a strong area of research in the medical field, and this has been caused by the high number of cases of brain tumors reported all over the world. The incidence rate of brain tumors in India in 2023 was found to be more than 38,000 new cases. This is one of the widespread cancers, with fewer than 34,000 deaths occurring. The opposing issues in the classification of brain tumors using machine learning (ML) include the small scale of the data, feature selection, and significant classification errors. Although deep learning (DL) methods are continually evolving, this research will provide solutions using a hybrid scheme that combines mixed ML and DL methods. The models examined in this paper are: Convolutional Neural Networks (CNN), CNN with Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and EfficientNetB0. Four types of measures were taken to establish the accuracy of the results and to compare the five types of models applied in the study. The outcome indicated that EfficientNetB0 was the best model, with an accuracy of 99.80%, a precision of 99.90%, a recall of 99.80%, and an F1-score of 99.80%. The CNN, CNN-SVM, CNN-RF, and CNN-LR models, on the other hand, provided accuracies in the range of 89% to 91%. This outcome demonstrates that EfficientNetB0 outperforms other models in detecting brain tumors, as well as its ability to enhance diagnostic capabilities in clinical settings. The paper presented the performance of each tumour class (Glioma, Meningioma, Pituitary Tumor, No Tumor) in each model. It also employed statistical validation methods, including cross-validation and significance testing, to support the results.