It is truly a challenge to detect a network phenomenon with an unpredictable persona. Due to the simultaneous dependency on network traffic and end users, events such as Flash Crowds are hard to predict. This paper introduces a Flash Crowd (FC) prediction methodology to operate at the edges of the ISP network hosting the hot-spot (i.e. end-server). Such a methodology would act as the logic unit in the detection architecture that we also propose. The proposed methodology promotes prediction with the use of a mathematical relationship between the request and response rate subject to the assumption that a FC is composed as a linear state model.