The invention provides a machine learning device that can optimize an equation for estimating a thermal displacement amount of a machine element based on the operating state of the machine element. The device acquires operating state data of the machine element, acquires a measured value for the thermal displacement amount, stores the operating state data and measured values for the thermal displacement amount of the machine element as training data associated together as a label, sets an equation for calculation the thermal displacement amount based on the operating state data, calculates anestimated value for the thermal displacement amount by substituting the operating state data into this equation, determines whether a difference between the estimated value for the thermal displacement amount and the measured value for the thermal displacement amount is no more than a predetermined threshold.