Improved Stroke Classification Accuracy by Using Hybrid Inception and Xception Models

Abstract

Purpose: Stroke is one of the leading causes of death and disability in the world that requires a fast and accurate diagnosis system. A major challenge in classifying strokes using deep learning is data imbalances, where the number of stroke patients is much less than that of non-stroke patients.

Methods/Study design/approach: This research proposes a Hybrid model approach that combines Inception and Xception architectures, and applies Synthetic Minority Over-sampling Technique (SMOTE) to balance the data distribution. The dataset used consisted of 5,110 entries with 12 stroke risk features, and evaluation was performed using accuracy, precision, recall, and F1-score metrics.

Result/Findings: The results show that the Hybrid model provides the best performance with an accuracy of 92.2%, outperforming the Inception (86.28%) and Xception (89.26%) models. In addition, the Hybrid model showed high and balanced precision and recall values, reflecting its reliability in detecting stroke cases.

Novelty/Originality/Value: The novelty of this research lies in combining the multi-scale feature extraction power of the Inception architecture and the depthwise separable convolution efficiency of the Xception architecture in a hybrid model. This approach is proven to excel in tabular data-based stroke classification and has the potential to be applied in automated medical diagnosis systems.