Meta-Analysis of Machine Learning Methods for Fruit Quality Prediction

Abstract

Considering the intentions towards diminishing fruit spoilage and the rising need for fruit spoilage detections; many of the models have been suggested which has been possible with the current rise of machine intelligence and computer vision. Nonetheless, the effectiveness of these suggested models is controversial once disclosed to unseen datasets and their adaptability is unsure when faced with diverse fruit types. The benefaction of this paper is to identify the ideal model for classifying defective fruits using a unified dataset and modifications to the existing models for better fruit spoilage detection for real-life implementations. Machine learning models like Support Vector Machine (SVM), Artificial Neural Networks (ANNs), and Convolutional Neural Networks (CNNs) are examined and fairly trained along with their processing phases, whereas their performances were estimated and analyzed in both binary and multiclass classification issues. Consequently, after the proposed modifications CNN-based models are the perfect solution. The suggested modifications to the CNN architecture improve the classification accuracy, precision, recall, and F1 score metrics. The experiment results show that the CNN1 method outperformed the other four state-of-art compared models in the prediction of fruit quality.