7. UNIVARIATE MODEL FORECASTING › 7.4 Analytic Technique Tutorial › 7.4.2 Processing Historical Observations

7.4.2 Processing Historical Observations

Often the series of historical observations present in a
capacity planning database file poses a number of problems
to the analyst attempting to develop forecasting models.
Among these problems are the following:

o  The presence of errant observations in the historical data
   series.  These observations result from holidays, missing
   data, and other anomalies.

o  Apparent random variations around a central trend.  This
   type of variation is common in random arrival workloads,
   such as testing or in a service bureau's workload.

o  Exponential trends associated with service level
   measurements.  Rather than being linear, the trends
   associated with service measurements are often exponential
   since they result from an underlying queuing relationship.

o  Step functions and other changes introduced by the
   addition of new workloads to the system.

The following sections discuss approaches to these problems:

 1 - Deleting Errant Observations
 2 - Data Smoothing Using a Geometric Moving Average
 3 - Data Smoothing Using a LOG10 Function
 4 - User-Defined Data Transformations