Outlier Removal for the Reliable Condition Monitoring of Telecommunication Services

Customer contentment plays an essential (if not the most important) role for the continuation and development of business relations. In this context, the functioning of goods and services acts as the crucial influencing factor. In the case of services, their providers willingly deploy CMSs (Condition Monitoring Systems) for the continuous CM (Condition Monitoring) of the service availability by means of various KPIs (Key Performance Indicators). A CMS must red-flag an abnormal condition. This happens if the recent value(s) of a KPI exceed(s) a predetermined threshold for a certain period. The pertinent literature contains a multiplicity of ways for automatic threshold computation, including a particular one for telecommunication services with time-varying load characteristic. Regrettably, the latter lacks in (distribution-independent) outlier extinction. Thus, this scholarly piece bridges this gap by applying an outlier detection algorithm based upon Walsh's nonparametric tests to a KPI history, removing the identified outliers, and comparing Pukelsheim's three sigma rule and the minimum or maximum value of the outlier-free array for threshold evaluation. The result of a corresponding field test assesses the reliability of the suggested methodology.