<![CDATA[Fat-filled powder has the potential to substitute milk in meeting the nutritional needs of the community, but its product quality remains unstable during continuous production processes. A key challenge in fat-filled powder (FFP) production is the difficulty in quality monitoring, which is influenced by various uncertainty factors that affect product quality. Machine learning can be implemented for quality monitoring system, but the imbalanced data conditions require the development of algorithms with optimal performance. This study aims to design a quality monitoring system for FFP using a machine learning model under imbalanced dataset conditions and the influence of other uncertainty factors. A Random Forest (RF) machine learning model was developed for monitoring FFP quality. In the context of imbalanced datasets, the model was optimized through various scenarios, including data splitting for training and testing, as well as the Synthetic Minority Oversampling Technique (SMOTE) and Distribution Optimally Balanced – Stratified Cross Validation (DOB-SCV) schemes. The results showed that the SMOTE model achieved the best performance in terms of accuracy, precision, and recall with scores of 99.67%, 99.79%, and 99.24%, respectively, on the testing data. Statistically, the RF model with the SMOTE data manipulation scenario also showed significant differences compared to the DOB-SCV model and the traditional data splitting approach. The quality monitoring model for FFP developed in this study can be implemented in the dairy industry, offering more stable, accurate quality monitoring predictions that align with real conditions, helping to avoid quality uncertainties during the production process. The implementation of this model in the industry has the potential to facilitate a broader, more transparent, and optimized product quality evaluation process, which can also be conducted in real time under continuous production conditions.]]>
Copyrights © 2025
