Recently, resistance element welding (REW), a method for welding dissimilar materials is gaining popularity in the automobile industry. However, in a three-sheet REW process, in which a rivet is inserted into the middle sheet, stable welding quality cannot be guaranteed because the alignment of the rivet and electrodes cannot be visually inspected. In this study, the three-sheet REW of SPFC590 (structural steel), SABC1470 (high-strength steel), and Al6061 (aluminum alloy) was investigated. The welding quality was evaluated at an increasing misalignment, and it was compared to that of a 0-mm misalignment distance. The welding quality was significantly reduced from a 6-mm misalignment distance. The pre-contact process, which is designed for predicting misalignment, was applied prior to the actual REW process. The dynamic resistance waveform in the pre-contact process was significantly different at each misalignment distance condition because of the difference in each alignment with electrodes and workpieces. The features extracted from the dynamic resistance waveforms were applied to a deep neural network (DNN) model to monitor the misalignment between electrodes and a rivet. Because the weld quality (weld nugget diameter) was not satisfactory at distances greater than 6 mm, the threshold value of the misalignment distance was set as 6 mm. A neural network that can predict the state of the misalignment between electrodes and a rivet was designed and trained, and the misalignment was predicted with 100% accuracy from DNN model.