Abstract | ||
---|---|---|
Several applications of supervised learning involve the prediction of multiple continuous target variables from a dataset. In the case of the target variables exhibit statistical dependencies among them, a multi-target modelling permits to improve the predictive accuracy of a regressor. Apart from describing the dependencies among the targets, the multi-target approaches could offer better performance and less overfit than traditional single-target methods. A group of multi-target regression methods have already addressed this demand with success. However, by modifying the Multi-Target Regressor Stacking (MTRS) from a single layer to a deep structure, we improved the prediction capability, and by an asynchronous tracking we could exploit the dependency of each target individually. This paper presents the novel method Deep Structure for Tracking Asynchronous Regressor Stack (DSTARS) and compares its performance to MTRS, Ensemble of Regressor Chain (ERC) and the independent regression by single-target (ST). As base-learners were used Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost) and Classification and Regression Tree (CART) to investigate the prediction capability of the different family of machine learning algorithms. The results carried out on ten diverse datasets show that proposed method achieved improvements over the others. |
Year | DOI | Venue |
---|---|---|
2017 | 10.1109/BRACIS.2017.30 | 2017 Brazilian Conference on Intelligent Systems (BRACIS) |
Keywords | Field | DocType |
machine learning,multi-target,multi-output,regression analysis | Decision tree,Asynchronous communication,Regression,Computer science,Support vector machine,Supervised learning,Artificial intelligence,Overfitting,Random forest,Machine learning,Gradient boosting | Conference |
ISBN | Citations | PageRank |
978-1-5386-2408-1 | 1 | 0.37 |
References | Authors | |
10 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Saulo Martiello Mastelini | 1 | 4 | 3.82 |
Everton Jose Santana | 2 | 3 | 2.11 |
Ricardo Cerri | 3 | 132 | 16.88 |
Sylvio Barbon | 4 | 46 | 10.97 |