An intelligent algorithm for final product demand forecasting in pharmaceutical units

Abstract

Accurate demand forecasting in pharmaceutical industries has always been one of the main concerns of planning managers because a lot of downstream supply chain activities depend on the amount of final product demand. In the current study, a five-step intelligent algorithm is presented based on data mining and neural network techniques to forecast demand in pharmaceutical industries. The main idea of the proposed approach is clustering samples and developing separate neural network models for each cluster. Using the obtained data, the performance of the proposed approach was assessed in a pharmaceutical factory. The optimal number of clusters for this case was four. Mean arctangent absolute percentage error, average relative variance, and correlation coefficient (R) were used to evaluate the performance of different neural network structures. The results of performing the models once for all data and once for the data of each single cluster showed that the forecasting error significantly decreased thanks to using this approach. Furthermore, the results indicated that clustering products not only raises the prediction accuracy but also enables a more reliable assessment of forecasted values for each single cluster. Such analyses are very important and useful for managers of marketing and planning departments in pharmaceutical units.