Drug Risk Classification Model Based On Machine Learning

Modern medicine and western medicine not only bring benefits to human health,but also bring endless fear and disaster to human life.Drug safety has always been the focus of attention in the world because of its characteristics of "treating diseases and causing diseases".Therefore,the risk monitoring of marketed medicines has been an important work of pharmacovigilance and drug risk assessment has been a hot topic in the medical community.Many foreign countries classify medicines into prescription medicine(RX)and over-thecounter(OTC)medicine on the basis of their safety and efficacy.While in China,they are divided into three categories: RX,OTC A and OTC B.The risk level of RX is higher than that of OTC,while the risk level of OTC A is higher than that of OTC B.The conversion of listed drug classes in the world is based on the application of drug manufacturers and the risk assessment of drug regulatory authorities to determine that the drugs that meet relevant national standards are OTC.The research on the conversion of RX and OTC has important practical significance since it is conducive to ensuring people’s medicine safety,improving people’s selfcare awareness,establishing a good medicine classification management system,and promoting the pharmaceutical industry to be in line with the international standards.According to the research,China mainly relies on experts’ experience in the evaluation process of conversion between RX and OTC,and lacks automated risk assessment model.Therefore,this article proposes to use the Adverse Drug reactions(ADR)Spontaneous Reporting System(SRS)database to objectively evaluate the medicine risk.To be specific,with the research data in the reports of ADR monitoring reports in our country and the using the support vector machine(SVM)classification in the machine learning technology,it is achievable to design a drugs risk evaluation index system and to realize automatic classification recognition.This model can provide technical support for the conversion of listed RX and OTC,and help to make decision for pharmacovigilance in China.The research content mainly includes the following aspects:1.Data collection and processing: conducted research and data collection on the SRS database of adverse drug reactions and the open catalogue of non-prescription drugs in China.From the data of ADR monitoring reports from 2010 to 2011 provided by the China Food and Drug Administration(CFDA),"western medicine" reports were selected as the research object.Then the database of drugs with categorization labels and the database of ADRs of "western medicines" were established respectively,and the data were preprocessed by standardization and integration.2.Construction and calculation of risk indicators: the relevant factors affecting drug risks were analyzed,and the deficiencies in the current research were pointed out.At present,the existing literatures mainly classify adverse reactions,and only consider the influence of a single factor on ADRs.In this paper,for the first time,Drugs were taken as the research object,and the drug risk assessment index system were proposed from the perspective of ADRs.Otherwise,three indicators for drug risk assessment were designed and defined: ADR severe reporting rate,ADR injury index and ADR coverage rate.A research dataset was established with the drug as the object,the risk index value as the feature,and the drug categories(RX,OTC-A and OTC-B)as the labels.3.Research on the application of multi-classification Support Vector Machine(SVM)algorithm: a multi-classification SVM algorithm was implemented by MATLAB and applied to the above research dataset.The accuracy of the model was improved by selecting the kernel function suitable for data characteristics and optimizing the penalty factor(c)and radial basis kernel parameter(g).The experimental results showed that the accuracy of the classifier model based on multi-classification SVM could reach 84.00%.4.Research on classification of class imbalance data: due to the unbalanced distribution of three types of sample data(RX accounts for 82.43%,OTC-A for 14.51%,and OTC-B for 3.06%),although the classification model based on SVM achieves a high accuracy,it was not ideal for the classification of OTC-A and OTC-B since they had smaller proportions.Therefore,before the classification,the class unbalance sampling technique was introduced to balance the number of different types.By using the Synthetic Minority Oversampling Technique(SMOTE),the sample sizes of small sample types were expanded and the classification model was reconstructed.The experimental results showed that the accuracy of the new classification model was 88.74%,of which the accuracy of RX was 88.14%,OTC-A was 90.47% and OTC-B was 96.77%.Based on the SRS reports of China,this paper constructs a drug risk assessment system based on ADR and establishes a drug risk classification model based on the SMOTE sampling technique and SVM.The experimental results show that this model has a high classification accuracy,which provides an automatic identification method for the conversion of RX,OTC-A and OTC-B drugs in China.