Mixed-Mode Tree

From Displayr
Jump to navigation Jump to search

A mixed-mode tree is a tree-based model where rather than have an outcome variable, as with a regression tree, one or more Variable Sets are used as the outcome.

In Displayr, mixed-mode trees are run by selecting Anything > Advanced Analysis > Machine Learning > Mixed-Mode Tree.

Detailed technical information can be found on the Q Wiki. The two programs are identical, except that Q can analyze more data and has more advanced options.

The target data is described in each of the boxes (called nodes).  The node shown at the 'top' of the upside-down tree shows the results for all the data. In this example, it shows that the target data is called Estimated profit to the industry and its average value is $6,192.

Each of the boxes shown underneath the tree is a segment of people from the database that have been found to be different in terms of their profit and related data.  Reading from the left to right we can see that:

  • If the people are working in a company with 48 or more employees which is in agriculture, forestry, fishing, communications services, transport or storage, their average profitability was around $61,000, which is roughly ten times the average for the total database.  Thus, this segment should likely be targeted.  However, note that it only accounts for 1% of the database.
  • Where there are 48 or more employees in one of the other industries, the profit drops to $18,621, which is a lot lower than the first segment but is still triple the average, suggesting that obtaining such firms as new customers is likely to be desirable (i.e., this is a prediction that comes from this analysis).
  • Where the number of employees is not known (i.e., 'Missing data'), or between 10 and 24, the average profit is $6,255.40.
  • Etc.

Predictive trees - 1.png