PENERAPAN HYPERPARAMETER TUNING UNTUK OPTIMASI CNN PADA PERANCANGAN APLIKASI IDENTIFIKASI PENYAKIT CABAI

Research Object: This study focuses on the application of hyperparameter tuning to optimize Convolutional Neural Network (CNN) in the design of a disease identification application for chili plants.Research Problem: CNN often achieves low accuracy and tends to overfit. In a previous study conducted by Dwi Suci Anggraeni, it was found that the CNN method achieved poor accuracy. This was due to the uncontrolled condition of the sample data and the limited dataset, which caused overfitting and low accuracy. Therefore, this study needs to improve the accuracy of CNN through hyperparameter optimization to reduce the likelihood of overfitting.Research Method: The method used is CNN with MobileNetV3 architecture, and hyperparameter tuning using grid search and random search. The data is taken from the Kaggle site, which contains images of chili plant diseases with five class labels. The research stages include data collection, preprocessing, model training, hyperparameter tuning, and model evaluation.Research Results: The results showed that random search provided the highest accuracy of 82%, while grid search resulted in an accuracy of 80%. The designed application using the best model was converted to TensorFlow Lite format and implemented in the Kotlin programming language with the Jetpack Compose framework.Implications of Research Results: This Android-based chili disease identification application is expected to help farmers quickly and accurately identify and handle disease and pest attacks on chili plants. The use of hyperparameter tuning has proven effective in improving the accuracy of the CNN model, making it adoptable in similar studies to enhance model performance in various plant disease detection applications.