<![CDATA[Manual fuel reporting in Indonesian logistics companies, such as PT Balaraja Distribusindoraya, often leads to inefficiency, fraud, and lack of anomaly supervision. This research aims to develop a web-based system that integrates machine learning and computer vision to monitor fuel consumption and detect anomalies in logistics fleets. The proposed system employs Isolation Forest for unsupervised anomaly detection based on fuel volume, travel distance, and fuel ratio, combined with a deep learning–based CBIR module using MobileNetV2 to validate fuel station images, and OCR to extract numerical data from receipts. Following the CRISP-DM methodology, the model was trained and deployed through a Flask-based API and evaluated using black-box and white-box testing. Experimental results show that Isolation Forest achieves the highest anomaly detection performance (F1-Score = 0.81, ROC-AUC = 0.99), CBIR validates official fuel stations with ≥95% similarity, and OCR reaches 97% accuracy in receipt recognition. The novelty of this study lies in its hybrid integration of anomaly detection and visual verification within a single scalable platform. This research contributes to Informatics by providing a framework for hybrid anomaly detection systems that enhance digitalization, transparency, and operational efficiency in the logistics sector.]]>
Copyrights © 2025
