Comments on an old paper about different classifiers?

Deep Learning
1

Hi,

I’m currently learning about classifiers.

Our dear professor linked us this paper from 2014: http://jmlr.org/papers/volume15/delgado14a/delgado14a.pdf

Which claims that Support Vector Machines is the most efficient way of classifying these datasets - even compared to neural networks.

I’m thinking that this paper from 2014 might be outdated, but haven’t yet found a recent comparison.

I realize the point that if the size of the training data is small then generative learning methods might be more efficient.

So, I’m curious. Does anyone have any recent result comparing any of these datasets using modern technologies compared to SVM and such?

(Unfortunately the exact link to the datasets mentioned in the paper [http://archive.ics.uci.edu/ml/datasets.html?task=cla] is not available anymore.)

Thanks in advance!

Best regards,
Aki

2

Completely forgot about this, but the dataset was actually still available at http://persoal.citius.usc.es/manuel.fernandez.delgado/papers/jmlr/data.tar.gz and I tested this out about year ago.

I threw in a dead simple model using sklearn’s MLPClassifier. Indeed, it seems that a simple modern NN easily beats most of the earlier results - at least in the cases where there is enough data.

I didn’t do anything complex. Just kept it super simple to see how it works out.

The datasets were really nicely organized and easy to use. If anyone happens to be interested on the subject, here is the notebook with results: https://gist.github.com/ikanher/0bd6b01e9b7fe35052c2db3710f7f1bc