Using Forecast Pro’s Built-in Outlier Detection and Correction

tipsandtricks

If you are forecasting a time series that contains an outlier, there is a danger that the outlier could have a significant impact on the forecast. One solution to this problem is to screen the historical data for outliers and replace them with more typical values prior to generating the forecasts. This process is referred to as outlier detection and correction.

Correcting for a severe outlier will often improve the forecast; however, if the outlier is not truly severe, correcting for it may do more harm than good. When you correct an outlier, you are rewriting the history to be smoother than it actually was and this will change the forecasts and narrow the confidence limits. This will result in poor forecasts and unrealistic confidence limits if the correction was not necessary.

A strong recommendation is that outlier correction should be performed sparingly and that detected outliers should be individually reviewed by the forecaster to determine whether a correction is appropriate.

Three Modes

Forecast Pro has three modes for outlier handling.

1) None instructs Forecast Pro to generate the forecasts using the actual (uncorrected) history. This is Forecast Pro’s default mode, and for many users this approach is appropriate.

2) Detection only instructs Forecast Pro to detect outliers and then present summary reports, listing each individual outlier with its suggested correction. In this mode, the forecasts are still generated using the uncorrected history.

3) Detection and correction instructs Forecast Pro to detect outliers and automatically use the corrected history when generating forecasts.

outliers tab

The Outlier settings tab in Forecast Pro TRAC allows you to specify your global approach to outlier handling. The Outlier settings tab is accessed by selecting Settings>Options on the main menu.

How it Works

Forecast Pro’s detection and correction algorithm works as follows:

  1. If the size of the largest error exceeds the outlier threshold, the point is flagged as an outlier and the historic value for the period is replaced with the fitted value.
  2. The procedure is then repeated using the corrected history until either no outliers are detected or the specified maximum number of iterations is reached.

You can adjust the Sensitivity setting to make the outlier threshold more or less sensitive. The proper setting will depend on the stability of your data.

An Example

As stated above, a strong recommendation is that outlier correction should be performed sparingly and that detected outliers should be individually reviewed by the forecaster to determine whether a correction is appropriate. In the following example we have selected “Detection only” mode on the Outliers tab under Settings>Options and then generated a forecast uing Expert Selection.

The next step is to review Forecast Pro’s outlier report. The screenshot below shows a tabular report listing all outliers that have been detected. In addition to tabular format, the information can be formatted as a presentation-style report. Outlier reports can be viewed on-screen or written directly to Excel.

Detected

This shows a Forecast Pro tabular outlier report including a graph. The report lists an outlier along with the suggested correction–the delta (difference) between the actual and the corrected. Note: the current status is that outliers have been “Detected.” No corrections have been made because the mode for outlier handling has been set to “Detection only.”

Notice that the graph is displaying both the actual history for May 2012 and the suggested correction. The data point is rather unusual and you can see that the forecast has a small spike in May 2013 which is being driven by the outlier.

Fine-Tuning

The next step is to review each outlier. Although Forecast Pro has a global setting for outlier handling (in this case “Detection only”), the software allows for individual outlier handling on an item-by-item basis.

Right clicking on an item in the Navigator brings up the Navigator’s context menu as shown in the screenshot below. The Outliers option on the menu provides four options available for outlier handling for that particular item, allowing you to fine tune your approach. The options are:

  • Default – apply the global outlier handling approach to the item.
  • Off – turn off outlier handling for the item and neither detect nor correct.
  • Detect – detect but do not correct outliers for the item.
  • Correct – detect and correct outliers for the item.

The Result

Correcting this outlier will change the shape of the forecast. The small spike in the April forecast will be easy to eliminate and the forecast will be smoother. The approach has been fine-tuned for this item.

corrected

This displays the forecast after Forecast Pro has “corrected” for the outlier. Notice on the Navigator that a modifier indicates that an outlier found for SKU VA-20-01 has been corrected. Compare the corrected and uncorrected screenshots above, and notice that correcting the outlier changed the forecasted values.

If you would like to see how Forecast Pro TRAC can help manage outliers and address other forecasting challenges, schedule a personalized Web-based demo with one our specialists.

One thought on “Using Forecast Pro’s Built-in Outlier Detection and Correction

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s