Intelligent System for Predicting Freeze-Drying Parameters for Camel Milk Powder Production Using Sensors and Machine Learning

Abstract

Producing high-quality camel milk powder requires precise control of vacuum sublimation drying parameters due to the product's sensitivity to thermal stress and oxidation. This study addresses the optimization of the drying process by integrating advanced machine learning (ML) techniques to predict optimal drying conditions. In this research, four ML algorithms were developed and evaluated: Logistic Regression, Random Forest, Decision Tree, and Extreme Gradient Boosting (XGBoost). Each algorithm was trained on detailed sensor data, including chamber temperatures, vacuum pressure, and milk thickness, with drying success defined by maintaining temperature deviations within ±5°C and completing drying cycles within 24 hours. The XGBoost model exhibited the best performance, achieving an accuracy of 99.3%, precision and recall of 96.4%, and the highest F1-score. Temperature parameters, particularly in specific chamber locations, emerged as critical predictors of successful drying outcomes. By enabling accurate forecasting and real-time parameter adjustments, this ML-driven approach significantly enhances drying efficiency, product quality, and sustainability, offering substantial economic and logistical benefits, particularly for remote regions in Kazakhstan where camel milk production is prominent.