Optimizing Drying Processes with Machine Learning: A Data-Driven Classification Approach

Abstract

Efficient drying of medicinal and agricultural plants is critical for enhancing food security and maintaining product quality during storage. This study investigates the application of advanced machine learning models—XGBoost, Polynomial SVM (Poly-SVM), and Radial Basis Function SVM (RBF-SVM)—to classify the drying status of five medicinal plants: Moringa, Neem, Lemongrass, Mint, and Hibiscus. The models were trained and tested independently for each plant type using a dataset of 35,000 experimental trials, with environmental parameters such as Solar Radiation, Wind Speed, Altitude, Humidity, and Temperature serving as inputs. Performance was evaluated using key metrics including Accuracy, Precision, Recall, F1-Score, and ROC-AUC.
The results show that XGBoost achieved the highest mean accuracy of 78.0% across all plant types, alongside superior precision (0.80) and ROC-AUC (0.73), making it highly effective in minimizing false positives. Poly-SVM demonstrated the strongest recall (0.98), effectively identifying optimal drying statuses, though with slightly higher false positive rates. RBF-SVM performed competitively with a mean accuracy of 77.8% but showed slightly lower boundary discrimination. These findings confirm that machine learning models can significantly enhance drying efficiency, contributing to food security by minimizing post-harvest losses and extending the shelf-life of agricultural products. Future research will explore real-time monitoring and feature optimization to further improve classification reliability.