The Impact of Hyperparameter Tuning in Convolutional Neural Network on Image Classification Model: A Case Study of Plant Disease Detection

Image Classification is one of the most popular uses of artificial intelligence. Deep Learning methodology with Con- volutional Neural Networks is found to be one of, if not the most accurate, way to classify images, with Xception and DenseNet architecture at the forefront of the performance metrics. Yet, the accuracy of those same models can still be improved by using an optimum hyperparameter tuning, namely, the AiSara tuning algorithm. This research contributes significantly to the field by assessing the impact of hyperparameter tuning on deep learning models in detecting various plant diseases. The datasets used are PlantVillage and PlantDoc, each representing a different image structure for the algorithm to train and test. Importantly, this research demonstrates that hyperparameter tuning on both CNN architectures achieved approximately a 2- 3% increase in accuracy compared to normal iterations. These findings highlight the practical utility of hyperparameter tuning in improving disease detection models and advance the field of image classification in agriculture.