Conservative Forecast Based On Historic Data

You can always use mean on the historic data as your forecast, which is reasonable. However, this is radical, especially when not consider the probability distribution. To be conservative, you may need to add a standard deviation (σ):

Mean + σ
Or
Mean - σ

In general, 1 standard deviation is good enough. In normally distributed data, 1σ means the side coverage of 34.1%, accounts for 84.1% of total population.

If there is no noise data within the dataset, arithmetic mean (average) can be used. However, if there is noise data, interquartile mean (IQM) is recommended. Accordingly, the standard deviation should be also based on the interquartile range (middle 50%).

In MySQL, the built-in function is STD(expr). In Oracle, it is STDDEV(expr) and in T-SQL, it is STDEVP(expr).

http://en.wikipedia.org/wiki/Standard_deviation
http://koncordpartners.blogspot.com/2010/11/practical-mean.html

http://oreilly.com/catalog/sqlnut/chapter/ch04.html
http://www.devguru.com/technologies/t-sql/7097.asp

No comments:

Post a Comment

Labels