An Empirical Comparison of Supervised Learning Algorithms
Rich Caruana, Alexandru Niculescu-Mizil

I replicated some of the comparison experiments performed by Caruana and Niculescu-Mizil which are described in their paper "An Empirical Comparison of Supervised Learning Algorithms". The original paper can be found here

I also gave a presentation of the paper on 04/29/2009. My presentation can be downloaded from here and code from here


Rich Caruana and Alexandru Niculescu-Mizil performed many experiments comparing different supervised learning algorithms over several data sets using 8 different performance metrics. These experiments are important because they provide a basis for direct comparison of different learning algorithms. Looking at their results, it is easy to see which algorithms are superior and which algorithms are best avoided.

This is not the first comprehensive empirical comparison to be performed. The STATLOG project (King et al) performed in 1995 was a European project which exhaustively compared different learning algorithms. However several new learning algorithms such as ANNs and random forests have been developed since then and so the STATLOG project is now outdated. Caruana and Niculescu-Mizil have performed a very comprehensive comparison of these newer algorithms.

To continue, Please click on one of the following links.

More information on the original paper
Experiments performed by me
Results of my experiments
Conclusions
Varun Sudhakar's home page