Optimasi Hyperparameter CNN dengan Arsitektur VGG16 Menggunakan Grid Search Untuk Klasifikasi Penyakit Buah Delima

Abstract

Early detection of pomegranate fruit diseases is crucial to reduce yield losses and improve harvest quality; however, visual identification in the field is often subjective and difficult due to the similarity of symptoms among different diseases. This study aims to develop a pomegranate fruit disease classification model using a Convolutional Neural Network (CNN) based on the VGG16 architecture, optimized through the Grid Search method. The dataset consists of five classes: 886 Alternaria samples, 116 Anthracnose samples, 966 Bacterial Blight samples, 631 Cercospora samples, and 1,450 Healthy samples, resulting in a total of 5,099 images. The dataset underwent preprocessing and data augmentation to increase variability and prevent overfitting. After balancing the dataset, it was split into 70% training data, 20% validation data, and 10% testing data. Hyperparameters such as epoch, batch size, learning rate, and optimizer were evaluated using Grid Search to determine the optimal configuration. The results indicate that the best performance was achieved using 100 epochs, a batch size of 32, a learning rate of 0.0001, and the Adam optimizer. The proposed model achieved a testing accuracy of 99.59%, with precision, recall, and F1-score values of 0.996. These findings demonstrate that the optimized VGG16-based CNN model is highly effective in accurately classifying pomegranate fruit diseases.