CLASSIFICATION OF RICE GRAIN VARIETIES WITH ENHANCED WILKS BASED PCA AND ENSEMBLE BASED LEARNING

Abstract

One of the most extensively produced grain crops in the world, rice, has several genetic variants. These categories differ from each other due to certain characteristics. These characteristics typically include color, texture, and shape. These characteristics that set distinct varieties of rice apart allow for the classification and evaluation of seed quality. The five varieties of rice that are most frequently grown worldwide are Basmati, Jasmine, Ipsala, Doongara, Wehani, Arborio, and Karacadag. The collection has 1,75,000 grain samples, with 25,000 of each category. A second set of 121 attributes was used, comprising 17 morphological features, 9 form, and 95 color features extracted from these images. The rice dataset was classified using an ensemble learning technique and a dimensionality reduction framework. This technique lowers the misclassification rate by aggregating the predictions of multiple classifiers. To ascertain the recall, prediction, and accuracy, statistical computations were performed. The classification accuracy of this model was 100% with XGBoost, 99.63% with Adaboost, and 99.85% using gradient boosting. The findings demonstrate the applicability of the models employed in the rice variety classification study.