Title: Machine learning in forecasting time series to control the stock of high-cost drugs used in cancer treatment

Authors: Paulo César Ossani; Lorena Enami Mazia; Gislaine Camila Lapasini Leal

Addresses: Department of Statistics, Exact Science Centre, State University of Maringá, Brazil ' Department of Production Engineering, Technological Centre, State University of Maringá, Brazil ' Department of Production Engineering, Technological Centre, State University of Maringá, Brazil

Abstract: An incorrect forecast can have tragic consequences for patient care. From a practical point of view, a financial compromise can be made as long as service levels are maintained at a maximum. Aiming at a compromise monetary reduction while still meeting demand, this study addresses a cancer drug inventory control in a small hospital. While there is extensive use of parametric time series models, this paper proposes the use of machine learning techniques, such as random forests, support vector regression, K-nearest neighbours, and extreme learning machines, which can produce better results in predicting drug consumption - monthly data between 2015 and 2019. Indicators such as inventory turns and service levels were used to validate the models, which increased inventory turns by more than 45%, which represents a large impact. Thus, better demand accuracy is achieved without compromising the service levels, proving the feasibility of the techniques.

Keywords: decision making; healthcare; k-nearest neighbours; machine learning; turnover.

DOI: 10.1504/IJSOM.2026.155906

International Journal of Services and Operations Management, 2026 Vol.54 No.4, pp.491 - 506

Received: 09 Aug 2023
Accepted: 23 Jan 2024
Published online: 26 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article