To improve the quality of additively manufactured parts, it is crucial to develop real-time process monitoring systems and data-driven predictive models. While various sensor- and image-based process monitoring methods have been developed to improve the quality of additively manufactured parts, very limited research has been conducted to predict surface roughness. To fill this gap, this paper presents a decision-level data fusion approach to predicting surface roughness in the fused deposition modeling (FDM) process. The predictive models are trained by the random forests method using multiple sensor signals. A decision-level data fusion method is introduced to integrate sensor data sources. Experimental results have shown that the decision-level data fusion approach can predict surface roughness in FDM with high accuracy.