Comparative Analysis of TimeGPT, Time-LLM and MSET Models and Methods for Transport Telematics
Boris S. Subbotin, P. I. Smirnov, Ekaterina Aleksandrovna Karelina, N. V. Solovyov, Vera V. Silakova · 2025
In this study, a comparative analysis of the TimeGPT, Time-LLM, and traditional MSET methods was performed on car telematics data, including such parameters as the degree of gas pedal pressure, speed, and steering wheel deflection. Forecast accuracy was assessed using MSE, MAPE and R2 metrics. The results showed that TimeGPT significantly outperforms MSET in accuracy, with MSE 0.023 vs. 0.035 for MSET, and MAPE 4.5% vs. 7.2%, respectively. The Time-LLM model also demonstrated good results, although it was inferior to TimeGPT in terms of accuracy. The data processing time for the TimeGPT and Time-LLM models was slightly higher, however, this is justified by their higher accuracy. The implementation of these models in transport systems can significantly increase safety and efficiency, optimizing routes and improving the prediction of vehicle behavior. The results confirm the potential of using modern models of time series analysis to improve the management of transport systems.