A PSO algorithm for improving multi-view classification

The Multi-view or multi-modality learning approach is becoming popular for providing different representations of a problem from which classifiers can learn from. Examples of these representations are, for instance, sound and image for the case of the video classification problem. The main idea behind multi-view learning is that learning from these representations separately can lead to better gains than merging them into a single dataset. In the same way as ensembles combine results from different classifiers, the outputs given by classifiers in different views have to be combined in order to provide a final class for an example. This paper proposes a PSO algorithm to combine the outputs coming from different views. It also considers that some views may be better at classifying specific classes, and provides weighting schemes for both views and classes. Experiments were performed in two datasets with three views each, and compared with all views in a single dataset, a majority voting scheme and a scheme based on the Dempster-Shafer theory. Experimental results show that the PSO obtains statistically better results than the other approaches evaluated.