HYBRID ARCHITECTURE FOR BRAIN STROKE PREDICTION USING ENSEMBLE TECHNIQUE

ABSTRACT This research presents an advanced ensemble deep learning model that combines the strengths of Convolutional Neural Network (CNN), ResNet50, and InceptionV3 architectures for accurate stroke prediction and diagnosis. The primary objective of the study was to enhance the reliability and accuracy of medical image-based stroke detection by leveraging the complementary capabilities of multiple neural networks. The CNN component efficiently extracted low-level image features, while ResNet50 contributed to overcoming vanishing gradient issues through residual learning, and InceptionV3 provided a multi-scale feature extraction capability. Together, these models formed an ensemble system capable of robust feature learning and improved classification performance compared to individual models. The methodology involved preprocessing medical images, training each base model separately, and integrating their outputs using an ensemble strategy that averaged or weighted predictions for optimal accuracy. The models were evaluated using key performance metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrated that the ensemble model consistently outperformed standalone architectures, showing superior generalization and reduced overfitting. The integration of diverse deep learning techniques allowed for enhanced feature representation, which contributed to higher sensitivity and specificity in stroke classification tasks. This study underscores the potential of ensemble learning in medical image analysis and stroke prediction. The combined CNN-ResNet50-InceptionV3 framework demonstrates that merging multiple deep learning architectures can significantly improve diagnostic accuracy and reliability. This research contributes to the growing body of work supporting AI-driven healthcare systems and highlights the importance of hybrid models for complex medical tasks. The proposed ensemble model not only achieves high performance but also sets a foundation for future implementations in real-time diagnostic applications and clinical decision support systems.