A graph-based approach to feature selection

In many data analysis tasks, one is often confronted with very high dimensional data. The feature selection problem is essentially a combinatorial optimization problem which is computationally expensive. To overcome this problem it is frequently assumed either that features independently influence the class variable or do so only involving pairwise feature interaction. To tackle this problem, we propose an algorithm consisting of three phases, namely, i) it first constructs a graph in which each node corresponds to each feature, and each edge has a weight corresponding to mutual information (MI) between features connected by that edge, ii) then perform dominant set clustering to select a highly coherent set of features, iii) further selects features based on a new measure called multidimensional interaction information (MII). The advantage of MII is that it can consider third or higher order feature interaction. By the help of dominant set clustering, which separates features into clusters in advance, thereby allows us to limit the search space for higher order interactions. Experimental results demonstrate the effectiveness of our feature selection method on a number of standard data-sets. © 2011 Springer-Verlag Berlin Heidelberg.