CLASSIFICATION OF PUMPKIN SEEDS USING MACHINE LEARNING TECHNIQUES

Abstract

Accurate and effective seed classification techniques are crucial for seed quality control and crop production optimization, as the need for healthy, high-quality seeds in agriculture continues to rise. With their high oil content and excellent nutritional value, pumpkin seeds are one of the main oil crops. A key component of precision breeding and variety enhancement is the identification and gathering of various pumpkin germplasm resources. Due to its sufficient amounts of protein, fat, carbohydrates, and minerals, pumpkin seeds are eaten raw, roasted, marinated, and sweetened as a dessert around the world. Thus, "UrğüpSivrisi" and "Çerçevelik," the two most significant and high-quality varieties of pumpkin seeds, which are often grown in Turkey's Ügrüp and Karacaören region, were the subject of this study. Nevertheless, measurements of 2500 morphological seeds of both types were achievable through the use of threshold approaches in their gray and binary forms. In order to identify the most effective technique for categorizing pumpkin seed varieties, all the data were modeled using six different machine learning techniques that took morphological features into account: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), kNearest Neighbor (k-NN), Decision Tree Classifier (DT), and Naive Bayes Algorithm (NV). A total of 87.06 percent for LR, 88 percent for SVM, 88.2 percent for RF, 87 percent for k-NN, 87 percent for DT and 86 for NV were the classifiers’ accuracy rates. The results have demonstrated that the proposed Random Forest classification Algorithm achieved a satisfactory overall accuracy of 88.2.