d
WE ARE EXPERTS IN TECHNOLOGY

Let’s Work Together

n

StatusNeo

How to Determine Forecastability of Demand?

In demand forecasting, there are a lot of cases where data scientists struggle to reach a good forecast error. On these occasions, they find it difficult to provide clear explanation to why some threshold of error is un-attainable no matter what technique is used. Forecast error of a product / event highly depends on its forecastability. If some event is unforecastable, it doesn’t even make sense to create a forecast model for it. We can rely on two parameters in order to determine whether the event is forecastable or not:

  • ADI (Average Demand Interval) – Represents the measure of intermittency.
ADI = Total Number of Period / Number of non-zero demand occasions
  • CV2 (Squared Coefficient of Variation) – Represents the measure of variation excluding zero demand events.
CV2 = Standard Deviation of Population /  Average of Population 

Based on these two parameters, we can classify the demand profile into one of these 4 different classes:

  • Smooth Demand
  • Erratic Demand
  • Intermitted Demand
  • Lumpy Demand

Smooth Demand

ADI < 1.32 & CV2 < 0.49 is classified as Smooth Demand because the demand is very regular in magnitude and occurrence. This class is very easy to forecast and low forecast error is attainable.

Sample Smooth Demand

Erratic Demand

ADI < 1.32 & CV² >= 0.49 is classified as Erratic Demand because the demand is irregular in magnitude even-though it is in regular occurrence. Low forecast error will be hard to attain.

Sample Erratic Demand

Intermittent Demand

ADI >= 1.32 & CV² < 0.49 is classified as Intermittent Demand because the demand is very regular in magnitude but it is very irregular in occurrence. Low forecast error will be hard to attain.

Sample Intermittent Demand

Lumpy Demand

ADI >= 1.32 & CV² >= 0.49 is classified as Lumpy demand because the demand is very irregular in magnitude and occurrence. It is improbable to create a forecast model that can have some predictability. Lumpy demand is unforecastable.

Sample Lumpy Demand

Determining the classification of historical demand makes it easy to plan out the research and development of a demand forecast model. This includes but no restricted to:

  • Prioritization
  • Algorithm selection
  • Setting baseline for acceptable forecast error

Reference : https://www.ias.ac.in/article/fulltext/sadh/045/0051

A Data Scientist with proven track record of working closely with product owners, managers, front-end developers, data engineers, and data scientists to execute research/analysis with rapid prototyping.

Add Comment