Accurate performance prediction models in gas turbines are essential for diagnosis. In this paper, a data-driven modeling method is newly proposed for an aero gas turbine engine system with high prediction accuracy during dynamic operation by considering the time delay included in the measured temperature from a sensor. Modeling and analysis are performed using an aero gas turbine engine with rapid thrust change. The accuracy of the data-driven model trained using the measured temperature and the flow temperature calculated through physical relations is compared. The flow temperature is calculated from the measured temperature based on lumped system analysis. In data learning, the low-pressure shaft speed and low-pressure turbine total temperature are defined as input parameters, whilst engine net thrust, and fuel mass flow rate are defined as output ones. The results show that using the measured temperature for training without conversion adversely affects the accuracy of the data-driven model. The flow temperature converted from the measured values is reasonable in terms of thermodynamic cycle matching of the engine. It is confirmed that the accuracy of the data-driven model is significantly improved when the actual flow temperature is used for model training.