More context would always be better. However, there is no reason Loess couldn't outperform a RandomForest and an SVM. Are we talking about just a single feature? If so, Random Forests would degrade into a bunch of jaggy trees. And I would assume a properly tuned SVM-regression would actually look quite similar to a Loess. If we're talking about multiple features, how exactly did you use Loess?
Shea Parkes
- 3.3k
- 1
- 18
- 13