Modelling of the Spray-Drying Process of Pectin Concentrates Using an Ensemble of ML Models and Interpretable AI Methods
Year: 2026
Type: conference-paper
Abstract: This paper presents an approach to computer modelling and intelligent optimisation of the spray-drying process for pectin concentrates, employing an ensemble of machine learning (ML) methods and explainable artificial intelligence (XAI). Within the Industry 4.0 framework, the influence of four key control parameters-inlet and outlet temperatures, air flow rate, and solution feed rate-was investigated using a full factorial experimental design. To construct a predictive model for product yield, an ensemble of $\mathbf{1 2}$ ML algorithms was evaluated, with the gradient boosting model achieving the best performance ($\mathbf{R}^{\mathbf{2}}=$ 0.97, RMSE $=0.21 \mathrm{~g}$). Model interpretability was ensured through SHAP (SHapley Additive exPlanations) analysis, enabling a quantitative assessment of factor contributions. Using differential evolution optimisation, optimal process regimes were determined (inlet temperature $\mathbf{1 8 5 - 1 8 9}{ }^{\circ} \mathrm{C}$, solution feed rate $\mathbf{1 4 - 1 5 ~ m l} / \mathbf{m i n}$), leading to a $\mathbf{3 1. 5 \%}$ increase in product yield without degrading functional properties. The proposed ML-XAI framework offers a practical methodology for integration into industrial engineering systems, supporting the development of digital twins and intelligent process control.
Authors: Sh. Yo. Kholov, M. K. Arabov, Z. K. Mukhiddin
Language: English
Subfield: Food Science
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7165205109