Deep learning and machine learning approaches have rapidly evolved and are expected to be effectively utilized in the heating, ventilation, and air conditioning (HVAC) industry. These technologies contribute to energy savings and CO2 reduction by improving the operational efficiency of HVAC systems. This study discussed different modeling methods for centrifugal chillers using machine learning approaches. A model for predicting the power consumption of centrifugal chillers was developed and examined. The model was applied to a centrifugal chiller with a single compressor and refrigeration cycle to explore its characteristics. The model’s accuracy was evaluated by quantitatively using the root mean squared error (RMSE) and qualitatively, through visualization of mechanical characteristics. As a result, when training the model solely with operational data, it was difficult to reproduce physical characteristics due to insufficient data. Therefore, it was required to pre-train the model with general machinery information or simulation data to overcome this challenge. Subsequently, the proposed model was applied to a centrifugal chiller with dual compressors and refrigeration cycles. The results indicated that the modeling method that applying for the single-cycle centrifugal chiller could also be applied to the dual-cycle centrifugal chiller by refining the preprocessing of the measured values.