Integrated Machine Learning Models for Bakery Product Defect Prediction

Abstract

The paper discusses the development of a model for predicting the probability of occurrence of defects in bakery products using a set of input variables at different stages of the technological process. The model is based on the analysis of data including control variables, such as oven temperature and humidity, as well as disturbance variables characterizing the properties of flour, the dough preparation process and baking of products. Based on the results of the study, a GMM-based model was selected, which demonstrated the highest accuracy, with the achieved Precision and Recall values equal to 1.0 for the class of defective products, which indicates high correctness of forecasts. In terms of Log-Likelihood, the model demonstrated a large difference between the classes, which confirms its ability to accurately classify both defective and non-defective products. The proposed model is an effective tool for predicting defects and optimizing process parameters. It allows you to adjust control variables, such as temperature and humidity, to reduce the amount of defects, ensuring stability of product quality. The article also proposes different methods for adjusting the values of control variables based on historical data. This allows for optimization of the technological process and improvement of the quality of bakery products in real-time production conditions.