Statistical Profiling of Hybrid CNN-SVM Effectiveness

Explaining decisions taken by a deep neural network is a challenging task. Even though significant advances have been made in this context during the last two decades, it remains a hot research problem. Combining a deep neural network with a classic AI model can be helpful in the context of model explainability. However, it may reduce its discriminating power. This paper performs statistical profiling on the effectiveness of hybrid deep learning models: a hybrid model of convolutional neural networks and support vector machines is used as a use case. We assess its performance on some state-of-the-art Latin and Arabic handwritten datasets. The proposed statistical study focuses on the hybrid deep learning model’s ability to recognize handwritten characters and maintain its performance in recognition rates compared to state-of-art CNN and SVM architectures. Obtained results show that the hybrid neural network achieves better overall classification accuracy. The findings of this research prove the effectiveness of hybrid deep neural networks and draw new bridges toward their explainability.