Fast and Efficient Cavendish Banana Grade Classification using Random Forest Classifier with Synthetic Minority Oversampling Technique.

Cavendish banana is an important export product of many countries, while postharvest banana classification also impacts plantation income. The quality inspection standard classifies Cavendish banana into three groups based on fruit size as large, medium and small. Most factories classify bananas manually which is time-consuming and also prone to errors. To expedite and enhance this process, a new fast and reliable method is proposed for classifying Cavendish banana gradings. The dataset contained 415 records, which were divided into five classes: L1, L2, L3, L4, and Reject. This study proposes employing a Synthetic Minority Oversampling Technique (SMOTE) to address imbalanced data by synthesizing the minority class to generate new minority sample data. Then a Random Forest classifier based on fruit length and fruit diameter parameters is implemented to efficiently classify Cavendish banana grading. Performance of the proposed method was assessed using 10-fold cross-validation. Compared to other machine learning techniques, results revealed that our approach delivered excellent performance with highest classification accuracy of 97.88%. Accuracy of 95.38% was achieved for an unseen testing dataset, illustrating superior performance to the other methods. We finally demonstrate a web application of the Cavendish banana grade classification system that was developed from the proposed model. [ABSTRACT FROM AUTHOR]