Fraud Detection in Rambutan Classification Using Deep Learning

Abstract - Mislabelling defective or raw rambutans as fresh constitutes fraudulent practices which undermine consumer confidence, fairness of markets, and integrity of supply chains. In this paper, we present an AI-powered solution for the detection of fraud in the classification of rambutan fruit by using deep learning and image processing for automation. The model is based on MobileNetV2, a pre-trained Convolutional Neural Network (CNN), which has been fine-tuned to classify rambutan into three classes: Fresh, Defective, and Raw. Advanced data augmentation, transfer learning, and fraud detection algorithms confer robustness to the model. By preventing misclassification fraud in agricultural markets, the suggested solution significantly improves food safety, transparency, and quality assurance. The efficiency and accuracy of manual rambutan fruit quality assessment are limited, especially in large-scale fruit processing. Hence, the current study overcomes some limitations of the manual system by proposing an automated classification system for rambutan quality based on image processing and machine learning algorithms. A number of inefficiencies in the manual assessment motivated the formalization of the proposed automated solution. A dataset was collected, featuring images of rambutans at different quality stages, followed by preprocessing methods for image enhancement. An applied deep learning classification model was trained on the labelled dataset. The results demonstrate the effectiveness of the proposed system in accurately categorizing rambutans into the distinct quality levels while also being able to detect fraud in terms of quality representation. As the proposed automated approach eliminates hindrances found in the manual system, it will improve the efficiency and precision of the fruit sorting process and provides useful insights to the agricultural and food processing industries.