Transient filter generation using parzen window method

An important portion of GSM network alarms are transient alarms which are short-term and low priority alarms. High number of transient alarms causes alarm pollution which may cause not only the extra time spent by the technicians but also the delays in handling of more important alarms. In order to prevent transient alarm pollution, the most common solution is to use transient filters. Domain experts create static transient filters according to alarm history. This study presents a new, statistical learning approach which creates transient alarm filters automatically and updates them according to the changing conditions and observations of alarms in the GSM network. Parzen window method is applied on alarm history and cumulative density function of alarm life-time is calculated for each alarm type. These cumulative density functions are used for automated decision of identity of transient alarms and transient alarm filter parameters.