Authors:
Linden Smith
1
and
John Blake
1
;
2
Affiliations:
1
Department of Industrial Engineering, Dalhousie University, Halifax, NS, Canada
;
2
Centre of Innovation, Canadian Blood Services, Ottawa, ON, Canada
Keyword(s):
Change Point Detection, Synthetic Data, Forecasting.
Abstract:
This paper describes tools to detect and estimate demand shifts for platelet products, through inventory monitoring, following the implementation of pathogen reduction (PR) technology at a pilot site in the Canadian Blood Services (CBS) network. A Statistical Process Control (SPC) framework was constructed to detect change points in inventory signals. A discrete event simulation is used to generate synthetic data for the inventory monitoring process. Both traditional forecasting and machine learning techniques were used to increase sensitivity to change detection and reduce time to detection by supplying the SPC algorithm with projected data. Experiments were run on data representative of changes in demand experienced at the pilot production site. It was found that larger shifts in demand had a higher probability of detection and a lower time to detection. Changes in demand, with an effect on the system larger than 10%, were almost always detected. Detection time varies greatly depen
ding on the level of the demand shift. Typically, shifts greater than 25% have an average detection time of just over a week while shifts of less than 5% have an average detection time of up to 25 weeks.
(More)