Predictive models of beetroot solar drying process through machine learning algorithms

Precise modeling of the drying process permits to achieve three key objectives: (i) assessing material properties, (ii) characterizing the microstructure, and (iii) optimizing the drying process. Driven by recent advances in machine learning techniques, we employed a machine learning-based approach to investigate the drying process of beetroot in a conventional solar dryer. Experimental part of this study showed that the drying kinetics of beetroot slices were highly impacted by the temperature and the thickness of the slices. Generally, the duration required for drying decreased as temperature and thickness increased. In one hand, the effective diffusivity coefficient was varying in a range of 5.65 × 10−9 - 7.37 × 10−7 m2/s. In other hand, the activation energy was ranging from 83.33 to 99.14 kJ/mol. The average activation energy for beetroot slices was approximately 90.47 kJ/mol. Findings show that the moisture transportation mechanism is dominated by liquid diffusion. In the modeling part, our findings suggest that the Catboost model is the most accurate among the evaluated models, based on three metrics: coefficient of determination (R2), Mean Squared Error (MSE), and Mean Absolute Error (MAE). Catboost model shows the higher performance with a R2 of 99.99%, MSE of 3.15 × 10−6, and MAE of 0.02.