loading page

Feature Engineering using Queuing Theory:
  • Ousainou Darboe
Ousainou Darboe

Corresponding Author:odarboe@gmail.com

Author Profile



Context: The increase in the number of patients in hospital emergency departments has a major impact on the quality of the provided care, patient satisfaction and financial cost of resources. Overcrowded emergency rooms and limited hospital resources to handle the high inflow of patients reduce the quality of service that is legally required by patients. Patients and their guides, get frustrated in ED queues due to long waiting times and sometimes leave without treatment. Controlling all these conditions can constitute an added burden to the emergency health care providers. The combination of all these issues limits the availability of emergency resources required for quick emergency medical treatment. The ability to predict patient length of stay might be valuable in understanding and planning of ED processes and resource utilization.
Objectives : We propose to use queueing theory as a feature engineering technique to improve the predictive ability of online learning models when predicting patient length of stay in EDs.
Methods : To conduct our experiments, we use the Massive Online Analysis framework to investigate the impact of using queueing theory as a new relevant feature for predicting patient length of stay on the Söder hospital ED data. The experiments are evaluated on two independent groups of 15 subsets without replacement. The Wilcoxon rank sum test is used to test the hypothesis that adding this new feature does not have a significant effect on the predictive error of online classifiers.
Results : The proposed online learning model predicts patient length of stay with a mean prequential accuracy of 82% compared to 61% without using the model. Using the queuing theory model, we can observe that the inter-arrival rate of patients at Söder hospital is 6 minutes. The average patient queueing time is 2 hours, whilst the average service time is 2.4 hours. Patients above 60 years have an average length of stay of nearly five hours, whilst younger patients have an average length of stay of 4 hours 24 minutes.
Conclusions : Queuing theory could be used as a new relevant feature in optimizing the prediction of patient length of stay for EDs. Additionally, we identified that elderly ED patients above 60 years spend more time than younger patients at the Söder hospital ED. This information could assist ED managers in setting up a strategic path for this age group to reduce the overall time patients spend at the ED. A DSS tool is implemented to assist ED managers in observing patient information, queueing and service time. This information could assist ED managers in making a real-time decision concerning patient flow and resource utilization in EDs.