The automotive industry prioritizes reliability and minimizes downtime, leading to a rapid embrace of predictive maintenance to curtail unforeseen failures and associated expenses. Traditional reactive maintenance strategies often prove inadequate in accurately forecasting failures, resulting in costly operational disruptions. This paper introduces a novel machine learning-driven method for predicting engine fuel system failures using anomaly detection algorithms applied to telematics data. By analyzing intricate relationships between fuel system components and operational features, our approach aims to enhance vehicle reliability and reduce maintenance costs. We present a comprehensive methodology encompassing data acquisition, data preprocessing, and feature engineering, leveraging key features such as rail pressure deviation, fuel pump metering valve current, engine torque, and engine speed to establish a benchmark for healthy vehicle operation, all sampled at one-minute intervals. A critical element of our method is the "Medium Surface," which represents the average healthy state across these features; deviations from this surface indicate minor anomalies, enabling early intervention. Performance evaluation using accuracy, precision, recall, and F1-score metrics reveals significant improvements over traditional methods, with our results achieving 83% precision on training data and 90% on field tests. Crucially, the model provides a 30-day predictive lead time for proactive maintenance. Acknowledging challenges such as data quality, computational complexity, and architectural limitations, this study underscores the advantages of our innovative approach and identifies promising avenues for future research in predictive maintenance within the automotive sector, ultimately contributing to increased efficiency and reduced operational costs.