Explainability enhancements in Anaplan Prophet and MVLR

edited February 11 in Groups

Our Forecasting data science team has recently released new enhancements to Explainability in Anaplan Prophet and MVLR.

Previously, PlanIQ engine was able to identify lagging indicators among the related data. So, for example, if change in certain indicator, for example fuel price, would have a delayed impact of 1 month or 2 months from its occurrence, PlanIQ would automatically lag the related data and using feature selection pick the most suitable lag. In explainability output, the name of the related data feature would appear together with its lag. For example Assomption, a holiday in some European countries, would appear as "Related: Assomption lagged 1 week", meaning that this holiday has a lagged impact of 1 week.

With the most recent enhancement, PlanIQ can now also recognize leading indicators. The engine shifts the related data "forward" to identify any leading impact and then utilizes feature selection to choose the most suitable leading lag. In the case of a leading indicator, the explainability output for the mentioned holiday would appear as "Related: Assomption lagged -1 week," signifying that this holiday has a leading impact of 1 week.

In some instances, the same event may have both lagging and leading impacts, and you can observe this through the explainability feature. See the attached example below:

Historical data in this exercise is based on a public dataset located at - https://github.com/Olivier-FONTAINE/Workplace-Attendance-Forecasting/tree/main