Machine learning helps to build predictive models to identify possible deviations and is the basis for innovative solutions for adaptive and predictive process control. The paper examines examples of using various machine learning methods to optimize the dough preparation process, and assesses their impact on the quality of the final product and the efficiency of the entire production process. The sponge method of pre-dough preparation involves two-stage fermentation and is one of the traditional methods in the baking industry, which ensures a rich taste and high-quality texture of bread. The main variables of the dough preparation process have been expertly established, based on which a model for monitoring and optimizing the main variables of the dough preparation process, which are target ones, is formed. The dough preparation process includes determining the optimal values of the initial variables characterizing the final quality of the dough. To achieve target values, models are used that predict them based on the current values of the input variables, including control and disturbing variables. Regression tree models were developed to predict various parameters, which made it possible to accurately determine the relationship between input and output variables. The patternsearch method has shown high efficiency in problems with non-smooth functions, namely, visible phenomena characteristic of real technological processes. Having indicated the need to use local minimum values and exclude stability to achieve optimal indicators, the method has shown positive results at the stages of training and testing models. As a result, optimization of the technological parameters of the dough preparation model for an additional direct search method has been achieved, and significant improvements in the accuracy of the automated process control system have been achieved.