In modern manufacturing, artificial intelligence and data analytics techniques are frequently used and developed for various industrial applications. The final purpose of the smart-factory problem is to automatically control the process recipe so that the manufacturer can obtain the desired standard or yield of the production processes. A datadriven model predictive control (D-MPC) strategy is widely used to quantitatively improve control systems. In this study, we introduce a novel modeling and control method termed stochastic response modeling (SRM). The method performs properly when a target variable has zero-inflated, intermittent, and time-invariant patterns. The proposed method comprises two main concepts: stochastic transformation of the response variable and coefficient adjustment algorithms that can address current limitations in the manufacturing field. Results of industrial case studies demonstrate the efficacy of SRM, especially in terms of the robustness and usability of model-based control. We believe that the proposed method can optimize the overall manufacturing process, such that high-yield production is always possible.