Examination and forecasting of drug consumption based on recurrent deep learning

Abstract

The provision of pharmaceutical drugs in quantities appropriate to consumption is an important point in the pharmaceutical industry and storage of medicines, as the production of large quantities of unnecessary drugs leads to a longer storage of drugs. Meanwhile most medicines have a short shelf life. When the amount of production is less than required, this affects the satisfaction of the customer and the marketing of the drug. Time series analysis is the appropriate solution to this problem. Deep learning has been adapted for the purpose of time series analysis and a prediction of the required quantities drugs. A recurrent neural network with Long-Short Term Memory LSTM has been used by deep learning. The proposed methodology is based on the seasonal number of prescription required quantities with the number of quarters as indicators. The aim of the research is to forecast the drugs amount needed for one year. The proposed method is assessed using two types of evaluation. The first one is based on MSE and the visualization of the actual data and forecasted data. The proposed method has reached a low value of MSE and the visualization graph is semi-identical, whereas the second evaluation method compares the result of the proposed method with traditional forecasting method. Multiple linear regression is a traditional prediction method used with the data set, whose results are relatively good and promising compared to the results of the traditional method.