adapt event detect time vari poisson process alexand ihler jon hutchin padhraic smyth time seri count data gener context web access log freewai traffic monitor secur log build data measur aggreg behavior individu human be typic exhibit period time number scale daili weekli reflect rhythm underli human activ make data homogen time data corrupt number bursti period unusu behavior build event traffic accid data mine problem find extract anomal event made difficult element paper describ framework unsupervis learn context base time vari poisson process model account anomal event show paramet model learn count time seri statist estim techniqu demonstr util model dataset partial ground truth form event freewai traffic data build access data show model perform significantli probabilist threshold base techniqu describ model investig degre period data includ systemat dai week time dai effect make infer detect event popular level attend experiment result propos time vari poisson model robust accur framework adapt autonom learn separ unusu bursti event trace normal human activ 