Decomposition of Service Level Encoding for Anomaly Detection

Muspratt R., MƏMMƏDOV M.

20th Australasian Data Mining Conference, AusDM 2022, Western Sydney, Australia, 12 - 15 December 2022, vol.1741 CCIS, pp.192-204, (Full Text) identifier

  • Nəşrin Növü: Conference Paper / Full Text
  • Cild: 1741 CCIS
  • Doi nömrəsi: 10.1007/978-981-19-8746-5_14
  • Çap olunduğu şəhər: Western Sydney
  • Ölkə: Australia
  • Səhifə sayı: pp.192-204
  • Açar sözlər: Anomaly detection, Feature decomposition, Healthcare provider, Outlier thresholds
  • Açıq Arxiv Kolleksiyası: Konfrans Materialı
  • Adres: Bəli

Qısa məlumat

Application of anomaly thresholds to health provider service level data encoded in vector form can lead to unintended consequences in terms of output complexity and business appropriate interpretation. In this paper we show that specific analysis incorporating feature decomposition with the application of relative business knowledge prior to selecting anomaly thresholds is an effective strategy across multiple health provider disciplines for addressing this complexity. Cluster definitions of Modal, Specialised and Aberrant are introduced as a descriptive framework on which to interpret said feature decomposition and to aid threshold setting. This strategy furthers the work introduced in [1] as refinement to an existing anomaly detection scheme.