Optimizing Support Vector Machine for Avocado Ripeness Classification Using Moth Flame Optimization

Avocado is a fruit from Mexico and Central America that is widely distributed worldwide for production and consumption. In avocados, ripeness is crucial because it is the primary factor consumers consider, significantly influencing their purchasing decisions. The manual ripeness selection is inefficient and inconsistent, so the classification system is essential for determining ripeness due to its effectiveness and efficiency compared to manual selection. In this study, we aim to develop a model that can classify avocado ripeness using machine learning with optimization. The data consists of avocado images categorized into five ripeness stages: underripe, breaking, ripe (first stage), ripe (second stage), and overripe. We utilize a Support Vector Machine (SVM) for the classification. Instead of manually choosing the model’s hyperparameters, we use Moth Flame Optimization (MFO) to optimize the SVM hyperparameters. The MFO ensures that the proposed model has optimal performance. For the input of SVM, we extract the HSV, GLCM, and HOG and apply PCA to the data. In this study, we use three SVM kernels: RBF, polynomial, and sigmoid. The MFO finds the model’s hyperparameters based on kernel requirements, including C, gamma, degree, and coef0. The MFO-SVM obtains optimal performance with an accuracy of 82.55%, 82.68%, and 81.23% for SVM kernel RBF, polynomial, and sigmoid, respectively. The results show that our proposed model demonstrates adequate performance in identifying the ripeness levels of avocados. The MFO increases model performance on all evaluation metrics compared to the baseline model and can be an excellent strategy to improve model performance.