KM-MIC: an Improved Maximum Information Coefficient Based on K-Medoids Clustering

Yali Zhang,Pengjian Shang
DOI: https://doi.org/10.1016/j.cnsns.2022.106418
2022-01-01
Abstract:In order to measure whether and how things are related, statistical correlation analysis comes into being. Among them, Pearson coefficients, Spearman and Kendall coefficients are widely used, but these correlation analysis methods cannot detect a wide range of relationship types due to their own limitations. Therefore, in 2011, Reshef et al. introduced a new correlation analysis method, maximal information coefficient (MIC), but this method has low computational efficiency and does not give an optimal data division method. On this basis, we propose the maximum information coefficient based on K-Medoids clustering (KM-MIC), which combines the K-Medoids clustering algorithm to optimize the way of data partitioning, and can quickly calculate whether there is correlation between the data. And this method has two major characteristics, generality and equitability.
What problem does this paper attempt to address?