Abstract: Milling force is a critical indicator of machining stability and finished surface quality. To improve the accuracy of milling-force prediction, reduce the number of input variables, and simplify the prediction process, a hybrid deep-learning neural network that indirectly predicts milling forces using spindle and feed-axis current signals is proposed. Firstly, the torque-matching characteristics of the machine tool are analyzed to establish the theoretical relationships between the currents of the spindle and feed axes and the resulting milling forces. Secondly, features related to milling-force variations are extracted from the current signals, and a deep-learning module is employed to further learn these features. A reinforcement-learning module is subsequently integrated to optimize the model, enabling it to adapt to different cutting conditions and perform force prediction. Finally, experiments under various cutting conditions are carried out to validate the proposed method using both force and current data, and comparative analyses are conducted against traditional neural-network models. The results demonstrate that the proposed model achieves accurate milling-force prediction across a wide range of machining parameters.