Distinguish Crude and Sweated Chinese Herbal Medicine with Support Vector Machine and Random Forest Methods

“Sweating” is a traditional processing method of Chinese medicinal materials in place of production. Because this operation is time-consuming, “sweating” is often abandoned, which affects the quality of Chinese medicine. At present, there is no specific method for identification of crude and sweated herbs. In this research, we tried to explore a rapid and effective discrimination method, in order to provide a new means for the quality control of Chinese medicine. We collected 120 batches of data of crude and sweated Chinese medicinal material Dipsaci Radix by near infrared spectroscopy. Support vector machine (SVM) and random forest (RF) were applied to construct discriminant models respectively. For comparison, principal component analysis (PCA)-mahalanobis distance (MD) discriminant was performed. Three models could well classify the test samples with the same error rate 3.33%. However the mean error rates were 1.73% for SVM, 1.13% for RF, and 9.18% for PCA-MD. Meanwhile, RF spent 1.06 s, the shortest time, in computation. Therefore, the performance of RF model is the best for discrimination of crude and sweated Dipsaci Radix.