Comparison of machine learning classifiers: A case study of temperature alarms in a pharmaceutical supply chain

作者：

Highlights：

• The false temperature alarm problem typical of rule-based systems is addressed.

• 10 machine learning methods are compared against the legacy system Gradient boosting achieved the accuracy of 95.9% (vs 16.6% by the legacy system).

• Strong features were represented by the temperature deviation and cargo location.

• Built models retain the best prediction stability when used on the new customers.

摘要

•The false temperature alarm problem typical of rule-based systems is addressed.•10 machine learning methods are compared against the legacy system Gradient boosting achieved the accuracy of 95.9% (vs 16.6% by the legacy system).•Strong features were represented by the temperature deviation and cargo location.•Built models retain the best prediction stability when used on the new customers.

论文关键词：Cold supply chain,Pharmaceutical,Temperature alarm,Rule-based monitoring,Machine learning,Prediction

论文评审过程：Received 26 February 2020, Revised 10 December 2020, Accepted 1 March 2021, Available online 17 March 2021, Version of Record 18 March 2021.

论文官网地址：https://doi.org/10.1016/j.is.2021.101759