Reading Partial Dependence Plots in Driverless AI
- Predictive AI
- H2O Driverless AI
This lesson walks through the Partial Dependence Plot in Driverless AI's model interpretation tab, using a credit score model to show how the average predicted probability of a bad score shifts as a single feature changes. It walks through the interest rate feature, where a slightly higher rate corresponds to a higher predicted probability of a bad score, and the outstanding debt feature, where predictions worsen for applicants above 1500 dollars of debt. The lesson also points out that the plot is interactive, letting a viewer zoom into a portion of the chart by dragging a red selection line.
Key takeaways
- How the partial dependence plot shows average predicted probability as one feature's value changes
- Why applicants with an interest rate slightly above the norm show a higher predicted bad credit probability
- What the outstanding debt feature reveals once applicants pass roughly 1500 dollars owed
- How zooming into a portion of the plot by dragging a red selection line helps closer inspection
Start owning your AI