A Comparative Study of Hyperparameter Tuning in Deep Learning Models using Bayesian Optimization and XAI

Plant disease detection is a crucial step in improving the quantity and quality of farm products since many plant diseases that arise in rice crops reduce the production of agriculture and cause financial loss. The manual evaluation of plant health is tiring and time consuming. Combining optimization algorithms with deep learning architectures can lead to more effective and efficient plant disease detection, with improved accuracy and generalization capabilities. This paper presents a comparison study on hyperparameter tuning of deep learning architectures for early detection of rice plant diseases. The hyperparameter tuning was performed by incorporating Bayesian Optimization algorithm on CNN, ResNet, MobileNet, InceptionNet and RegNet. The experimental study is conducted on dataset containing 1007 images of rice seed crop that contains 501 images of healthy samples and 506 unhealthy samples. Performance evaluation metrics such as accuracy, precision and inference time are employed to compare models. Additionally, a comparative analysis using explainable Artificial Intelligence (XAI) is conducted to visualize the interpretability of hyperparameter combinations and their impact on disease detection.