View on GitHub

Bayesian Detection

The Bayesian approach to the problem of detecting changes in distributions, can be described in the following terms. Suppose that we decide to monitor the stability of a process with a statistic $T$, having a distribution with p.d.f. $f_T(t;\theta)$, where $\theta$ designates the parameters on which the distribution depends (process mean, variance, etc.). The statistic $T$ could be the mean, $\bar X$, of a random sample of size $n$; the sample standard-deviation, $S$, or the proportion defectives in the sample. A sample of size $n$ is drawn from the process at predetermined epochs. Let $T_i$ $(i = 1,2,\cdots)$ denote the monitoring statistic at the $i$-th epoch. Suppose that $m$ such samples were drawn and that the statistics $T_1,T_2,\cdots,T_m$ are independent. Let $\tau = 0,1,2,\cdots$ denote the location of the point of change in the process parameter $\theta_0$, to $\theta_1 = \theta_0 + \Delta$. $\tau$ is called the change-point of $\theta_0$. The event $\tau = 0$ signifies that all the $n$ samples have been drawn after the change-point. The event $\tau = i$, for $i = 1,\cdots,m-1$, signifies that the change-point occurred between the $i$-th and $(i+1)$-st sampling epoch. Finally, the event ${\tau = m^+}$ signifies that the change-point has not occurred before the first $m$ sampling epochs.