A robust deep learning ensemble framework for accurate brain tumor classification

Prognostication of brain tumors by developing highly optimized deep learning based convolutional neural networks is a significant area of research in medical informatics. The research has been focused on optimization of parameter learning mechanisms while significantly improving the accuracy of classification through varied innovative machine learning training and testing methodologies. The research presented in this paper utilizes an ensemble technique to develop a novel brain tumor classification model that demonstrates competitive performance in medical image analysis. An exhaustive experimentation is conducted to compare the prediction results of the ensemble model with five base convolutional neural network models as well as boosting-based convolutional neural network models named Adaptive Boosting and Extreme Gradient Boosting. Comparative analysis is performed using various prediction quality indicators such as accuracy, loss, and F1-scores. The proposed ensemble model achieves 97% accuracy for brain tumor detection, representing a 2–5% improvement over individual base models and demonstrating highly competitive performance compared to boosting technique-based convolutional neural network models. Furthermore, experimentation has been performed for parameter tuning of the optimizer function using six different optimizers. The results show highly accurate prediction levels for the suggested ensemble model that makes use of the Adam optimizer, achieving optimal convergence and classification performance.