Reviewing Automated Insights

Automated Insights Dialog Basics

WATCH A TUTORIAL:  https://lityx.com/automated-insights-tutorial/

When Automated Insights for a dataset are ready to review, you will find the results in the Automated Insights dialog.  This can be opened using either of the first two methods mentioned in https://support.lityxiq.com/074862-Setting-up-Automated-Insights.

When you open the dialog, if the insights are not ready yet, you will see a message explaining the reason (for example, they may still be in progress, or you may have never setup the insights for the dataset).

If they are ready, you will see the target variables listed in the drop down at the top of the dialog.  Your options will look like the following:

 

 

Target Variable to Analyze - select the target variable for which you wish to analyze Automated Insights. If a variable you had requested for automated insights does not appear here, it likely means there was an error computing insights for that variable.

Insight Type - You will see three options: Key Drivers, Multivariate Segmentation, and Segmentation Details.  Each is described in more detail below.

Go and Refresh Variables buttons - Click Go after changing either the variable or the insight type.  Click Refresh Variables if Automated Insights is still processing.  If it has since completed, you will see the variable list become populated.

 

Insight Type - Key Drivers

The Key Drivers insight provides an evaluation of the most important predictors of the selected target variable.  They are ranked according to relative importance, with 100 assigned to the variable found to be most predictive.  These insights should be interpreted as directional - a starting point for understanding and further exploration or deeper machine learning modeling.

 

Insight Type - Multivariate Segmentation

The Multivariate Segmentation insight type provides insight into how multiple variables combine to improve understanding of the target variable.  The output is viewed in the form of a tree.  The full dataset is the starting point at the top of the segmentation diagram.  The dataset is continuously broken into smaller segments based on the values of predictor variables.  Each "node" in the tree structure can be considered as representing a segment of the dataset, and is evaluated for its effect on the target variable.  See below for more information on evaluating the information provided.

 

 

Insight Type - Segmentation Details

The Segmentation Details insight can be thought of a tabular view of the graphical multivariate segmentation.  Each segment (node) in the multivariate segmentation is represented as a row in the Segmentation Details table, along with information on that segment.  The table includes a full description of the segment definition.  The rows can be sorted by clicking on column headings.  In particular, it can be interesting to sort the table based on the average value of the target variable to immediately see which segments tend to lead to strong (or weak!) results.  The Segment Size column provides information on how prevalent each segment is overall.