Studies comparing the effectiveness of models for drying bitter gourd slices

Drying is an essential food preservation method, improving product shelf life and quality while reducing transportation and storage costs. This study evaluated the drying kinetics of bitter gourd slices under halogen drying conditions using both traditional empirical models (Page, Midilli, Logarithmic, Peleg, and Two-Term) and the machine learning-based random forest (RF) model. Experiments were conducted at 60 °C, 65 °C, and 70 °C with slice thicknesses of 3, 5, and 7 mm. Model performance was assessed using the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute percentage error (MAPE). The results show that the RF model demonstrated the highest accuracy, with an average R2 of 0.9826, the lowest RMSE (0.0655), and MAPE (1.40 %). Its ability to capture non-linear drying behaviour made it the most reliable model. The Midilli model was the best-performing traditional model, with an average R2 of 0.9851, but its accuracy declined for thicker slices and higher temperatures. Logarithmic and Peleg models exhibited significant errors, particularly during the mid-to-late drying phases. The results highlight RF's robustness and adaptability, outperforming traditional models in handling complex drying dynamics.