Leveraging Lime for Trustworthy Apple Quality Assessment

This study explores the integration of machine learning (ML) models and explainable artificial ıntelligence (XAI) techniques to enhance the accuracy and transparency of apple quality assessment. Leveraging the Apple Quality dataset from Kaggle, we evaluated the performance of random forest (RF), logistic regression (LR), support vector machine (SVM), and eXtreme gradient boosting (XGBoost) across accuracy, precision, recall, and F1-score metrics. Among these models, RF demonstrated superior performance, achieving an accuracy of 87.6%, underscoring its robustness in managing complex feature interactions. A novel contribution of this study lies in employing the local ınterpretable model-agnostic explanations (LIME) method to elucidate model predictions, offering localised, human-readable insights into the factors influencing classification decisions. This approach bridges the gap between ML advancements and practical applications, ensuring that the results are accessible to non-technical stakeholders, such as farmers and supply chain managers. The findings highlight the potential of integrating ML with XAI in agricultural quality assessment to optimise resource utilisation, minimise post-harvest losses, and improve supply chain efficiency. Future work could address limitations by expanding the dataset to encompass diverse apple varieties and exploring more granular quality classifications.