Fitted Value vs. Observed Data
The Traditional Approach found significant differences between Vehicle Body: ‘Coupe’ and ‘UTE’. However when we look at the predictions from this model on the hold out data, we see they do not perform well. The Algorithmic fitted values are much closer to the observed data.
The same is also true when we look at rating area. The Traditional Approach found significant differences between Rating Areas ‘D’ and ‘F’. Again, the Algorithmic Approach produces more accurate predictions on hold out data. Similar trends were observed in the other benchmark tests but are not shown.