Research On Relevance Forecasting Methods Of Demand For Drug Based On Prescription Data

Along with the increase in varieties of diseases, the supply-demand relationship and circulation link of drug are becoming more and more complicated. At the same time, the use of drug quality is becoming extremely fluctuated. It casuses lots of undesirable inventory phenomenon. Sometimes the demand for drug exceeds the supply, and sometimes the suppply exceeds the demand. And the similar phenomenon is more and more frequent. So how to accurately grasp the rule of demand for drug, so as to predict the usage of drug is an urgent practical problem which needs to address for hospital managers. On the other hand, with the further development of Hospital information management, there exists accumulated large amounts of data about the usage of drug. These data is based on the pathology. It reflects the situation of patients, prescribing habits of doctors, the category and quantity of drug and the relationship among different drugs. How to use these data to find the knowledge, rules and patterns of medical compatibility and medicine dosage with data mining technology, then use of these knowledge and patterns to forecast the demand for drug, especially to forecast the demand based on the relevance among drugs, have been the difficult problem for hospital managers. It also has received widespread attention of domestic and foreign scholars, and become a research hotspot. This issue is put forward in this background, and it’s a bold attempt to forecast the demand for drug from simple time series forecasting to causality prediction combined with the internal relevance among drugs.This thesis focuses on how to forecast the demand for drug with data mining technology, based on the data gathered through field investigations. From the perspectives of both time sequence and relevance about the data, the content includes the following three aspects:Firstly, from the perspective of extracting time-sequence information from sample data, a predictive model of demand for drug is established based on auto regressive integrated moving average model(ARIMA). Then considering that ARIMA cannot deal with the nonlinear character of data and BP neural network model has excellent learning ability of the nonlinear relationship, the idea of combination forecasting model is introduced to establish an ARIMA-BP forecasting model to forecast the demand for drug. And the model is practical verified through example tests.Secondly, from the perspective of extracting relevance information from sample data, a predictive model of demand for drug is established, considering prescribing habits of doctors, pharmacological properties of drugs and demand rules of drugs. In order to accomplish the above work, firstly, a correlation analysis model of Apriori is established, based on prescription data of drug. Then a predictive model of demand for drug is established based on BP neural network forecasting model. Considering of some disadvantages of BP neural network model, genetic algorithm(GA) is introduced to optimize the initial weights and threshold values of BP neural network model to establish a GA-BP forecasting model. And the model is practical verified through example tests.Thirdly, from the perspective of extracting both time-sequence and relevance information from sample data, the idea of combination forecasting model is introduced to establish an intelligent non-linear combination forecast model based on GA-BP algorithm, in order to further improve the forecast accuracy of demand for drug. The model combines both ARIMA-BP and GA-BP forecasting model. Through example tests and model comparisons, it is verified that the model is practical and it is better than that two single forecasting models.