COD: online temporal clustering for outbreak detection

摘要: We present Cluster Onset Detection (COD), a novel algorithm to aid in detection of epidemic outbreaks. COD employs unsupervised learning techniques an online setting partition the population into subgroups, thus increasing ability make over as whole by decreasing signal-to-noise ratio. The method is adaptive and able alter its clustering real-time without need for detailed background knowledge population. attempts detect cluster made up primarily infected hosts. argue that this technique largely complementary existing methods outbreak can generally be combined with one or more them. show empirical results applying problem detecting worm attack on system networked computers, thIs approximately 40% lower infection rate at false positive 1 per week than best previously reported data set achieved using HMM model customized task.