An outlier is a data point that falls outside of the expected range of the data, i.e., it is an unusually large or small data point1. Outliers can have a significant adverse impact on the forecasts, as they can skew the data distribution and distort the statistical analysis2. Therefore, it is important to detect and remove outliers from the demand data before generating forecasts.

One of the techniques that can be used to detect outliers is to use the standard deviation of the data, or the equivalent z-score, to determine the outlier limit3. For example, one approach is to set the lower limit to three standard deviations below the mean, and the upper limit to three standard deviations above the mean. Any data point that falls outside this range is detected as an outlier.

However, detecting outliers is not enough. The most appropriate next step would be to screen the outlier for manual review. This means that the detected outlier should be examined by a humanexpert to determine whether it is a true outlier or not, and whether it should be corrected or not4. This is because not all outliers are erroneous or irrelevant. Some outliers may be valid observations that reflect real changes in demand, such as seasonal peaks, promotional effects, or market trends. In such cases, correcting or removing the outliers may lead to inaccurate or biased forecasts.

Therefore, screening the outlier for manual review can help verify the cause and validity of the outlier, and decide on the best course of action. Some of the possible actions are:

Correcting the outlier: replacing the outlier with a more typical value based on historical data or expert judgment. This can smooth out the data and reduce the noise.

Separating the demand streams: splitting the data into two or more series based on different factors that influence demand, such as product type, customer segment, or distribution channel. This can isolate the outliers and allow different forecasting methods to be applied to each series.

Adjusting the forecasting model: modifying the parameters or assumptions of the forecasting model to account for the outliers, such as using a different smoothing factor, trend component, or error term. This can improve the fit and accuracy of the model.

References: 1: Outlier Definition 1 2: How to Forecast Data Containing Outliers 2 3: How to Detect Outliers in Machine Learning – 4 Methods for Outlier Detection 1 4: How Outlier Detection and Correction Works 4 : How to Understand What is an Outlier in Forecasting 3