14.13 Biased Estimate from the Training Dataset

We noted above that evaluating a model on the training dataset on which the model was built will result in overly optimistic performance outcomes. We can compare the performance of the randomForest::randomForest() model on the te dataset presented above with that on the training dataset. As expected the performance on the training dataset is wildly optimistic. In fact it is common for the randomForest (Breiman et al. 2022) model to predict perfectly over the training dataset as we see from the confusion matrix.

predict_tr <- predict(model, newdata=ds[tr, vars], type="class")
con(predict_tr, actual_tr)
##       Predicted
## Actual No Yes
##    No  76   3
##    Yes 14   8

Similarly Figure @ref(fig:memplate:rf_riskchart_tr) illustrates the problem of evaluating a model based on the training dataset. Again we see perfect performance when we evaluate the model on the training dataset. The performance line (the Recall which is plotted as the green line) follows the best achievable which is the grey line.

pr_tr <- predict(model, newdata=ds[tr, vars], type="prob")[,2]
riskchart(pr_tr, actual_tr, risk_tr, title.size=14) +
  labs(title="Risk Chart - " %s+% mtype %s+% " - Training Dataset")


Breiman, Leo, Adele Cutler, Andy Liaw, and Matthew Wiener. 2022. randomForest: Breiman and Cutler’s Random Forests for Classification and Regression. https://www.stat.berkeley.edu/~breiman/RandomForests/.

Your donation will support ongoing availability and give you access to the PDF version of this book. Desktop Survival Guides include Data Science, GNU/Linux, and MLHub. Books available on Amazon include Data Mining with Rattle and Essentials of Data Science. Popular open source software includes rattle, wajig, and mlhub. Hosted by Togaware, a pioneer of free and open source software since 1984. Copyright © 1995-2022 Graham.Williams@togaware.com Creative Commons Attribution-ShareAlike 4.0