Classification of Apple Quality Based on Physical and Chemical Properties: A Machine Learning Based Approach

Abstract

This study examines a machine learning-based approach for assessing physical and chemical properties to determine apple quality. While traditional quality control methods are time-consuming, costly, and subjective, artificial intelligence and computer vision techniques offer faster and more accurate results. The study used a dataset consisting of 4000 samples containing key physical and chemical variables such as apple size, weight, sweetness, crispness, juiciness, ripeness, and acidity. The performances of various machine learning algorithms were compared during the training and testing phases. In model performance evaluations, the Voter Classifier (VT) algorithm achieved the highest accuracy rate of 91.25% and F1-Score 91.25% also demonstrated superiority in other key metrics. In the study conducted in Trabzon, the combination of the LGBM (Light Gradient Boosting Machine) and CatBoost algorithms within the voter structure stood out as an innovative approach that increased model performance. While this method has limited applications in the literature, it has the potential to make significant contributions to the optimization of quality control processes in the agricultural sector. Therefore, it is concluded that AI-supported systems are an effective tool for agricultural quality assessment.