Multi-target or multi-output regression is another instance of the more general learning task of multi-target prediction. Here the prediction targets are real-valued, as opposed to the closely related task of multi-label clasification where the target variables are binary. There are numerous applications for multi-target regression such as stock market prediction, power generation forecasting and ecological modeling.
Despite its importance, multi-target regression has received relatively small attention from the Machine Learning community. The main reason is that multi-target regression is usually casted into independent regression problems and tackled with single-target regression algorithms. Our main contribution is the development new multi-target regression methods, which are inspired by well-known multi-label classification methods and are able to successfully model target dependencies. A further contribution is the creation of several new multi-target regression data sets, which can be used as benchmarks in future evaluations.
Data sets and Software
The multi-target regression data sets that we collected and used in our evaluations can be found here.
The new multi-target regression methods and the evaluation framework have been implemented as an extension of Mulan. The code can currently be found in Mulan's SVN repository, while we plan a new release which will include several improvements in few weeks.
Please contact Eleftherios Spyromitros-Xioufis for questions, comments, suggestions or request for help with the source code and the data sets.
- E. Spyromitros-Xioufis, W. Groves, G. Tsoumakas, I. Vlahavas, "Drawing Parallels between Multi-Label Classification and Multi-Target Regression, Issues when Targets are used as Inputs to Meta-Models", under review in the Machine Learning Journal, 2013.
- E. Spyromitros-Xioufis, W. Groves, G. Tsoumakas, I. Vlahavas, "Multi-label Classification Methods for Multi-target Regression", arXiv preprint arXiv:1211.6581, 2012.