Advanced Driving Cycle Generation Using XGBoost: A Case Study in Iğdır, Türkiye

Accurate driving cycle models are crucial for evaluating vehicle performance and emissions, particularly in diverse urban environments. This study presents a novel framework for driving cycle generation using On-Board Diagnostics (OBD) data and the Extreme Gradient Boosting (XGBoost) algorithm. Focusing on the city of Iğdır, Turkey, a region characterized by complex and variable traffic conditions, we collected comprehensive driving data across 19 distinct cycles. Traditional driving cycle generation methods, such as Markov chains, often oversimplify the intricate driving behaviors observed in such environments. To address this, we employed XGBoost, which excels at modeling non-linear relationships and capturing temporal dependencies in large datasets. Our results demonstrate that XGBoost significantly outperforms traditional Markov chain-based models, achieving a Root Mean Squared Error (RMSE) of 0.3575 and a Mean Absolute Error (MAE) of 0.2474 on the test set, compared to an RMSE of 9.5621 and an MAE of 7.3322 achieved by the Markov chain model. Feature importance analysis revealed that temporal dependencies, particularly lagged acceleration and speed variables, were critical in accurately predicting vehicle behavior. The generated driving cycles effectively reflect the real-world driving patterns in Iğdır, offering a more precise tool for vehicle testing and emissions estimation. This study not only highlights the limitations of traditional methods in capturing complex driving patterns but also demonstrates the potential of machine learning, specifically XGBoost, in improving driving cycle generation. The framework developed in this research can be adapted to other regions with similar traffic challenges, contributing to more accurate and region-specific vehicle performance evaluations.