Health products sales forecasting using computational intelligence and adaptive neuro fuzzy inference systems

In our days the importance of reducing the inventory level in a healthcare organization is increasing fast. As a result, the value of an accurate supply forecast is becoming more relevant. The main objective of this paper is to analyze and compare some of the most popular and widely applied techniques available based on computational intelligence. In addition, it aims to demonstrate the competitiveness of the aforementioned using real‐world data. The methods that are employed include neural networks (feed forward, radial basis, generalized regression, and recurrent networks) and the hybrid neural fuzzy system (ANFIS). The experimental part of this study is conducted with the use of sales’ data extracted from the database of a major Greek medical supplier. A two‐year period was employed in order to gather the appropriate sales figures about some of the most popular medicines and subsequently each technique’s forecasting ability was tested against a third year. The parameters of each technique are then fine‐tuned in order to minimize the performance functions. Finally, a brief statistical analysis of the techniques used is performed to facilitate the comparison between them and define the most appropriate method for this particular issue.