Which Supervised Learning Method Works Best for What? An Empirical Comparison of Learning Methods and Metrics

Decision trees may be intelligible, but do they perform well enough that you’d really want to use them? Have SVMs replaced neural nets, or are neural nets still the best models for regression, and SVMs best for classification? Boosting maximizes a margin similar to SVMs, but can boosting compete with SVMs? And if it does compete, is it better to boost weak models, as theory might suggest, or to boost stronger models? Bagging is much simpler than boosting – how well does bagging stack up against boosting? Breiman said Random Forests are better than bagging and as good as boosting. Was he right? And what about old friends like logistic regression, KNN, and naive bayes? Should they be put out to pasture, or do they fill important niches?

In this talk we compare the performance of ten supervised learning methods on nine criteria: Accuracy, F-score, Lift, Precision/Recall Break-Even Point, Area under the ROC, Average Precision, Squared Error, Cross-Entropy, and Probability Calibration. The results show that no one learning method does it all, but some methods can be “repaired” so that they do very well across all performance metrics. In particular, we show how to obtain the best probabilities from max margin methods such as SVMs and boosting via Platt’s Method and isotonic regression. We then describe a new ensemble method that combines select models from these ten learning methods to yield much better performance. Although these ensembles perform extremely well, they are too complex for many applications. We’ll describe what we’re doing to try to fix that. Finally, if time permits, we’ll discuss how the nine performance metrics relate to each other, and which of them you probably should (or shouldn’t) use.

Speaker Details

Rich Caruana is an Assistant Professor in Computer Science at Cornell University. He received his Ph.D. at CMU in 1997 where he worked with Tom Mitchell and Herb Simon. Before joining the faculty at Cornell in 2001 was on the faculty at the Medical School at UCLA, and at CMU’s Center for Learning and Discovery (CALD). Rich’s research is in machine learning and data mining, and in the application of these to areas such medical informatics, ecology, and computer architecture design. He is perhaps best known for his work in inductive transfer, semi-supervised clustering, and optimizing learning for different performance criteria. Rich likes to mix algorithm development with applications work to insure that the new methods he develops actually work well in practice.

Date:
Speakers:
Rich Caruana
Affiliation:
Cornell University