Skip to Main content Skip to Navigation
Journal articles

Improved principal component analysis for anomaly detection: Application to an emergency department

Abstract : Monitoring of production systems, such as those in hospitals, is primordial for ensuring the best management and maintenance desired product quality. Detection of emergent abnormalities allows preemptive actions that can prevent more serious consequences. Principal component analysis (PCA)-based anomaly-detection approach has been used successfully for monitoring systems with highly correlated variables. However, conventional PCA-based detection indices, such as the Hotelling’s and the Q statistics, are ill suited to detect small abnormalities because they use only information from the most recent observations. Other multivariate statistical metrics, such as the multivariate cumulative sum (MCUSUM) control scheme, are more suitable for detection small anomalies. In this paper, a generic anomaly detection scheme based on PCA is proposed to monitor demands to an emergency department. In such a framework, the MCUSUM control chart is applied to the uncorrelated residuals obtained from the PCA model. The proposed PCA-based MCUSUM anomaly detection strategy is successfully applied to the practical data collected from the database of the pediatric emergency department in the Lille Regional Hospital Centre, France. The detection results evidence that the proposed method is more effective than the conventional PCA-based anomaly-detection methods.
Document type :
Journal articles
Complete list of metadata
Contributor : Mylène Delrue Connect in order to contact the contributor
Submitted on : Monday, November 15, 2021 - 9:20:23 AM
Last modification on : Tuesday, November 16, 2021 - 3:56:47 AM

Links full text




Fouzi Harrou, Farid Kadri, Sondès Chaabane, Christian Tahon, Ying Sun. Improved principal component analysis for anomaly detection: Application to an emergency department. Computers & Industrial Engineering, Elsevier, 2015, 88, pp.63-77. ⟨10.1016/j.cie.2015.06.020⟩. ⟨hal-03428177⟩



Record views