Friction stir blind riveting (FSBR) is a new process for joining automotive lightweight dissimilar materials. During FSBR, a blind rivet rotates at high speed, contacts with the upper sheet or workpiece of a lap joint, and then penetrates the workpieces. Using FSBR to join carbon fiber-reinforced polymer composite and aluminum alloy sheets has been studied experimentally, however, the quantitative relationship between the FSBR process and joint quality/strength remains unclear. To gain a better understanding of FSBR, the proposed method effectively models this relationship by integrating data de-noising, feature extraction, feature selection, and classifier fusion. Engineering-based features are extracted directly from the FSBR penetration force and torque signals; data-driven features are extracted using principal component analysis. Regression models and kernel support vector machines (SVMs) are trained and fused for quality prediction. The proposed method provides online monitoring of FSBR and prediction of joint quality.