Detecting novel multi-variable associations in big data based on MIC
Fubo Shao, Keping Li · 2015
It is meaningful to discover valuable relationships in big data. The maximal information coefficient (MIC), a new measure of dependence of relationships, was proposed by Reshef et al. in 2011, and an approximate algorithm was designed. But the algorithm designed by Reshef et al. (2011) can only calculate the MIC of two-variable relationships. In this paper, an algorithm (BKM-MIC) is proposed. To our best knowledge, the BKM-MIC algorithm is the first algorithm calculating the MIC of multi-variable relationships. And based on the BKM-MIC algorithm, a matrix iteration algorithm with pruning (MIP) is designed. A simple example shows that MIP algorithm can not only reduce computation workload, but also can precisely identify dependent and independent multi-variable relationships.