In order to quantify energy efficiency potentials of metal cutting machine tools, it is necessary to determine the time shares of different energy states. This paper presents a machine learning approach analyzing energy states, developed according to the acrfullCRISPDM, to improve the accuracy of the time study compared to static approaches. Different concepts, such as Convolutional Neural Network (CNN) or Long Short-Term Memory (LSTM), are deployed and evaluated based on electrical load profiles from an industrial use case with 35 metal cutting machine tools, where both approaches achieve an accuracy of over 95 %.