Intelligent brain tumor classification using a Bi-level optimized stacking ensemble with metaheuristic feature selection and Bayesian hyperparameters

Brain tumor is a major neurological disorder that may severely influence the living standards of an individual. Early and precise diagnosis of a patient determines the quality of life and prognosis with a lot of accuracy. The accuracy of Magnetic Resonance Imaging has become significant, as Artificial Intelligence automates the process and proves to be an important asset in medical imaging. Several studies have looked at different processes to classify the tumor, but there is still a big need for research into how to make a framework that improves all of the parts and combines the predictive abilities of different models. To fill this gap, this study proposes a novel hybrid stacked model for classifying brain tumors into multiple groups. The suggested SMOTE-based TransSecOpt framework starts with transfer learning to get the basic features, followed by Synthetic Minority Over-sampling Technique (SMOTE) to balance the data. Particle Swarm Optimization (PSO) was used for feature selection, and the Optuna framework was used to fine-tune the hyperparameters of several machine learning models. The best optimized models are used as base learners in a stacking. This results in a meta -classifier that effectively integrates the strengths of individual base learners. The suggested model exhibits outstanding performance, attaining a recall of 92.34 %, accuracy of 92.34 %, precision of 92.61 %, and F1-score of 92.37 %. The model’s robustness is corroborated by a Jaccard score of 86.00 %, a Cohen’s kappa score of 89.49 %, and a Matthew’s correlation coefficient of 89.58 %, demonstrating its efficacy as an automated brain tumor diagnosis tool.