Dec242001 Apr042002 Jul142002 Oct232002 Feb012003 May132003 Aug222003 Dec012003 Mar112004 Jun202004 Sep292004

figure 14.9 The same information as Figure 14.8,but over a longer period. It shows alerts based on changes from yesterday on the timescale of several years. The "comb" shape of the alarm levels indicates that the algorithm almost always signals an alarm on Mondays.

upper count as the moving average algorithm, which we will describe in the next section.

Using yesterday's count as the baseline has at least two problems, however. First, the baseline itself will be subject to random fluctuation, so if day t happens to have a low count, then day t+1 will automatically look more dangerous. Second, day-of-week effects are a disaster: Mondays will always signal an alert since they follow the very low-counting Sundays.

Was this article helpful?

0 0

Post a comment