CNN-ML Stacking for better Classification of Rice Leaf Diseases

Rice is still a very important commodity for society at large. For the people of Indonesia, rice is the main source of staple food every day. Rice disease is one of the things that needs to be handled immediately so as not to cause suboptimal harvest results. Several studies on rice disease classification have been conducted using machine learning models. However, the classification of rice diseases is still a challenge that must be resolved. In this study, the ensemble model is utilized to perform stacking on CNN and machine learning to perform better classification of rice leaf diseases. The CNN architectures used as first-level learner are MobileNetV2, DenseNet121, and NASNetMobile. Logistic Regression, SVM, KNN, Naïve Bayes, and Random Forest were selected as meta-learners. The dataset processed from various sources amounted to 440 images of rice leaf diseases divided into 5 classes. The results obtained show that the proposed model can improve the accuracy of classification performance to 97.5% when compared to the performance of three standard CNN models. Exploration of various CNN architectures and meta-learners is a research gap that can be done for further development.