Ensemble Diversity Measures Calculated Prequentially on Evolving Data Streams

The following small multiples present visualizations of prequentially calculated ensemble diversity measures on streams with various types of concept drift. Columns in each grid represent different datasets, whereas rows showcase different ensemble sizes (labels on the right). A detailed description of algorithm parameters and a short analysis of these results can be found in: Brzezinski, D. and Stefanowski, J., “Ensemble Diversity in Evolving Data Streams”.

1 Diversity measures for different ensemble sizes (Bag)

1.1 Accuracy

1.2 AUC

1.3 Q statistic

1.4 Disagreement

1.5 DoubleFault

1.6 KohaviWolpert

1.7 CFD

1.8 Kappa

2 Diversity measures for different ensemble sizes (AUE)

2.1 Accuracy

2.2 AUC

2.3 Q statistic

2.4 Disagreement

2.5 DoubleFault

2.6 KohaviWolpert

2.7 CFD

2.8 Kappa

3 Comparison of all diversity measures for k=10 components (Bag)

4 Comparison of all diversity measures for k=10 components (AUE)