A successful inventory management strategy requires a proper compromise between service levels to meet demand with the costs of storing products that lose quality or value with time. This research analyzes different approaches for estimating the safety stock of medical supplies using simulation tools and machine learning algorithms. The research was conducted in four stages: analysis, design, development, and validation. During the analysis stage, the number of patients coming to the emergency room over a 12-month period were collected and preprocessed. In the design stage, simulation tool, forecasting methods, machine learning algorithm, and comparison criteria were selected. In the development stage all necessary models were implemented and tested. Estimating safety stock for medical supplies using classical formulae has limitations since it does not consider variability. Instead, average number of patients in a given period is included. However, fluctuations may consume the safety stock, leading to stockouts. One alternative is to forecast the number of patients using 3-month moving averages instead of the annual average. Another approach is to apply machine learning techniques for forecasting. The simulation results showed that using safety stock based on average numbers can lead to as many as 10 stockouts per year. The model based on a 3-month moving average reduces this number to 6 annual stockouts, while the machine learning model further decreases the figure to only 2 stockouts per year. Although the results depend on the specific dataset used, the proposed models based on dynamic safety stock estimation might help reduce the occurrence of stockouts, thereby helping healthcare institution meet their service levels.

Keywords

Inventory Management, Security Stock, Service Level, Discrete Event Simulation, Machine Learning.