Chili Leaf Disease Classification Using Transfer Learning with VGG16 and MobileNetV2 Combined with Random Search Hyperparameter Tuning

Abstract

Chili is one of the main food commodities in Indonesia with considerable economic value. Frequent climate changes have made chili plants more vulnerable to pest and disease attacks. Early identification of these diseases is crucial, as delays can lead to crop failure. However, this process presents its own challenges, as it requires specific expertise and considerable time. This study employs the transfer learning method using the VGG16 and MobileNetV2 architectures to build a model capable of classifying diseases in chili plants based on leaf images, along with the use of Random Search hyperparameter tuning to improve model accuracy. The results show that the use of transfer learning for disease classification achieved high accuracy, with MobileNetV2 reaching an accuracy score of 88% without tuning. Meanwhile, the application of Random Search hyperparameter tuning proved effective in improving model accuracy, particularly with the VGG16 architecture, which saw a significant accuracy increase from 51% to 89%. It can be concluded that the transfer learning method is well-suited for identifying diseases in chili plants based on leaf images with high accuracy, and that the application of Random Search hyperparameter tuning successfully enhanced the model’s performance.