Rapid screening and classification of common drugs using Raman spectroscopy based on the hybrid and cascade-SVM model

In the field of drug detection, achieving rapid and accurate screening and classification of drugs and non-drugs is crucial. To meet this demand, this study proposes a new strategy that combines Raman spectroscopy technology with machine learning algorithms. Using a dataset of methamphetamine and heroin samples for model training, a Hybrid and Cascade-SVM model was constructed. This model consists of two OneClassSVM models and a standard SVM model. The unique hybrid cascade architecture design helps to enhance the model's discrimination accuracy. Although the model is trained only on the drug dataset, its performance is comprehensively evaluated during the validation and testing phases using a dataset that includes both drugs and non-drugs. The two OneClassSVM models first screen out drug samples from the mixed data, and the SVM model subsequently performs binary classification on these screened samples. Experimental results show that the Hybrid and Cascade-SVM model demonstrates good performance in drug identification and classification. This method provides an efficient solution for rapid on-site drug detection, optimizes the data processing workflow, and offers new insights for detecting drugs in complex matrices.