An Enhanced Rice Grain Classification Using a Novel Feature Engineering and Random Forest Modelling
This research presents a novel approach for rice grain classification, distinguishing between the Cammeo and Osmanick varieties using a dataset obtained from the UCI Machine Learning Repository. The dataset includes morphological features such as area, perimeter, major axis length, minor axis length, and eccentricity. To enhance predictive performance, a Fibonacci-derived linear and polynomial feature engineering technique was introduced, effectively increasing the dataset's feature space. The engineered features were combined and dimensionality reduction was applied using Principal Component Analysis (PCA) with 56 components. Feature selection was conducted using a Random Forest classifier with 500 estimators and a maximum depth of 15. Subsequently, a Random Forest model was employed for classification. The model achieved an accuracy of 93.70% and a mean cross-validation score of 91.34% (5-fold). Class-wise performance metrics for Cammeo included a precision of 0.92, recall of 0.95, and F1-score of 0.93, with support of 350 samples. For Osmanick, the precision was 0.95; recall 0.93, and F1-score 0.94, with support of 412 samples. The confusion matrix highlighted a strong classification performance, with 331 correct classifications for Cammeo and 383 for Osmanick. The dataset was obtained from https://www.muratkoklu.com/datasets/, it is named Rice Dataset (Commeo and Osmancik). This work demonstrates the efficacy of combining this method in enhancing the classification of rice grain varieties.
Download Paper