Machine Learning-Based Orange Quality Classification: A Hyperparameter Optimization Approach Through Puma Optimizer
Harun Akbulut, Sema Atasever, Eyüp Sıramkaya
Abstract
Oranges are used in various industries, including food, beverage, cosmetics, and pharmaceuticals, in addition to being consumed fresh. Traditional orange quality assessment methods are expensive and often lead to human errors. This study employs machine learning (ML) algorithms to significantly improve the accuracy of classifying orange quality. We conducted evaluations on multiple ML algorithms, specifically Random Forest (RF), XGBoost, Gradient Boosting (GB), and Decision Trees (DT). Among the evaluated methods, RF algorithm demonstrated the best performance, initially achieving an accuracy rate of 75.51%. To enhance model performance, we utilized the newly developed Puma Optimizer Algorithm (POA) for hyperparameter optimization. Following optimization, the accuracy of RF algorithm increased to 81.63%. Meanwhile, XGBoost and GB achieved an accuracy of 79.59%, and the DT model saw an improvement to 63.27%. The results were validated using a set of metrics such as accuracy, precision, recall, F1 score, along with correlation and confusion matrices. The enhancements in classification performance underscore the effectiveness of ML algorithms in practical agricultural applications.