Cylinder bore honing is a finishing process that generates a crosshatch pattern with alternate valleys and plateaus responsible for enhancing lubrication and preventing gas and oil leakage in the engine cylinder bore. The required functional surface in the cylinder bore is generated by a sequential honing process and is characterized by Rk roughness parameters (Rk, Rvk, Rpk, Mr1, Mr2). Predicting the desired surface roughness relies primarily on two techniques: (i) analytical models (AM) and (ii) machine learning (ML) models. Both of these techniques offer certain advantages and limitations. AM's are interpretable as they indicate distinct mapping relation between input variables and honed surface texture. However, AM's are usually based on simplified assumptions to ensure the traceability of multiple variables. Consequently, their prediction accuracy is adversely impacted when these assumptions are not satisfied. However, ML models accurately predict the surface texture but their prediction mechanism is challenging to interpret. Furthermore, the ML models' performance relies heavily on the representativeness of data employed in developing them. Thus, either prediction accuracy or model interpretability suffers when AM and ML models are implemented independently. This study proposes a hybrid model framework to incorporate the benefits of AM and ML simultaneously. In the hybrid model, an artificial neural network (ANN) compensates the AM by correcting its error. This retains the physical understanding built into the model while simultaneously enhancing the prediction accuracy. The proposed approach resulted in a hybrid model that significantly improved the prediction accuracy of the AM and additionally provided superior performance compared to independent ANN.