Surface roughness prediction studies in fused deposition modeling (FDM) process are usually based on a perimeter profile of each deposited layer. This study categorizes three types of build edge profile composed of perimeter, raster, and combination of both layer deposition patterns, which have been considered to reduce the predictive error of roughness models. Furthermore, an innovative approach based on combination of theoretical and empirical methods is used to analyze and predict the randomness in the geometry of build edge profiles. The same methodology is used to model the roughness profile and surface roughness behaviors. The proposed models have been tested for robustness against varying build orientations and with data available in the existing literature. The robustness of the proposed models is compared with the existing models. The results clearly demonstrate that the proposed models are very useful in reducing prediction errors.