/
Using Path Prediction Machine Learning

Using Path Prediction Machine Learning

Owned by Pete Trevitt
Last updated: Aug 30, 2022
3 min read

Note: This feature will only be visible once enabled and once a model has been built. For more information contact Alterian Support or your Account Manager.

The Journey Visualisation flow displays the stream of Event Locations visited and, when selecting one of the “Last N Days” filters, the display is based on the most recent data.

The Path Prediction feature allows that view to be extended into the future by seeing a display of predicted activity by Visitors over the next N number of events, based on previous activity, and predicting from the end of the current Visitor stream.

There are two stages to this process, as with any modelling based on Machine Learning

  1. Training the model - Training a model is an off-line process, likely to be run by a data analyst as part of an overnight process. This process may take several iterations and a review of the results, and re-training of the model may be required before the model is ready for use in the UI. This process is not covered in the User Help, and is best discussed with Alterian representatives as part of on-going discussion on the requirements.

  2. Path Prediction - Once the model has been trained and stored against the event stream in question, the Analyst can then use the Journey Visualisation flow diagram to display a number of steps into the future to allow a view of potential paths taken by Visitors based on their previous activity

Path Prediction

As defined in the stages above, using Path Prediction to visualisation potential future events, does rely on a model being trained and available. Once a model has been trained Prediction mode is available.

Selecting the Prediction Mode option displays a UI for configuration:

  • Available Queues - This dropdown will display queues that have a trained model against them. If this is not the case the UI will simple not display any and obviously not allow execution of a prediction.

  • Date Range - This option will display one of the “Last N Days” options. this is because Path Predictions can only be made from the last user interaction. NOTE: The date range here is used to create a list of Visitors (CXID) who have interacted with the brand in that defined period. The process then predicts using the entire visitor stream of those Visitors who have made that recent interaction. More recent data might be better used for prediction, but there is flexibility here to consider older streams.

  • Number of Events - This option allows the analyst to select the number of predicted events to display and these will be added to the right hand side of the current right aligned Sankey.

Once executed, new prediction are coloured blue to differentiate them from the last actual Visitor interaction in green. If any of the above configurations is not correct, the execution icon will ne disabled, with a tool tip.

IMPORTANT NOTES:

  • Execution of Path Prediction is a complex process that may take several minutes depending on data volumes. This process must allowed to finish uninterrupted.

 

 

 

 

Related content