Optimization of Banana Ripeness Classification Using Machine Learning and Deep Learning Approach Through Hyperparameter Tuning

Abstract Accurate determination of banana ripeness is essential in the fruit supply chain to ensure quality and reduce postharvest losses. This study aims to compare the performance of seven classification algorithms: support vector machine (SVM), eXtreme gradient boosting (XGBoost), convolutional neural network (CNN), artificial neural network (ANN), naïve Bayes (NB), random forest (RF), and k-nearest neighbors (kNN) in identifying banana ripeness based on digital images. The dataset consists of banana images at four ripeness levels, processed using two main approaches: HSV color feature extraction and image conversion to 1D for classical machine learning algorithms and direct image classification for CNN models. Each algorithm was optimized through hyperparameter tuning to achieve the best performance. According to the results, CNN emerged as the high performing algorithm, with an accuracy of 0.99. Another algorithm, such as ANN, RF, and XGBoost, also performed well, with an accuracy ≥ 0.95. In contrast, NB showed lower performance with an accuracy of 0.87. These findings highlight the strength of deep learning approaches in banana ripeness classification. The findings of this study serve as a foundation for the further development of automated fruit ripeness detection technologies.