Efficient MiningofDatathrough ReuseinaPublic Safety Network

Traditional statistical analysis ofnetwork dataisoften employed todetermine traffic distribution, to summarize patterns ofuserbehavior, ortopredict future network traffic. Mining ofnetwork data maybe usedtocharacterize userbehavior patterns, todiscover hidden usergroups, todetect payment fraud, or toidentify network abnormalities. We combine thistraditional traffic analysis withdatamining techniques andanalyze traffic data collected fromadeployed public safety trunked radio network. After data cleaning andtraffic extraction, weidentify clusters oftalk groups byapplying clustering algorithms onpatterns represented bythehourly number ofcalls. Traffic prediction models arethendeveloped by applying classical prediction models ontheaggregate andclustered data. Cluster-based prediction approaches, while less computationally demanding, perform wellcompared totheprediction based on theaggregate traffic.