This thesis presents an application of kernel-based methods for identification of the feed drive's dynamics model parameters and prediction of the disturbances affecting feed drives during operation from the cutting forces. To this purpose, the Partially Linear - Least Squares Support Vector Machine (PL-LSSVM) and Kernel Recursive Least Squares - Tracker (KRLS-T) algorithms were utilised for batch identification and online prediction of disturbances. Experimental case studies were performed with two ball screw feed drives under simulated and real cutting conditions to verify the identification and in-operation prediction of cutting forces.