Multi-Model Machine Learning Approaches for Fault and Anomaly Detection in Petroleum Product Distribution: A Comparison

Maintaining the integrity of petroleum transportation is essential for operational safety and efficiency. Substantial issues arise from faults in fuel transportation systems, including distribution valve malfunctions and anomalies in transit patterns, such as theft or unexpected fuel losses. This work presents a machine learning approach that combines supervised fault detection models, like Logistic Regression, SVM, Random Forest, and XGBoost, with unsupervised anomaly detection methods, such as k-means, Isolation Forest, and Autoencoders. The study employs real fuel transportation data and anomaly detection datasets obtained from petroleum distribution networks. Experimental results show that combining supervised and unsupervised methods leads to better detection accuracy and reliability, making the petroleum delivery system more dependable.