Maximal Information Coefficient And Its Application To Brain Network Analysis

With the development of statistical method, the research on its application to all kinds of fields goes deeper and deeper. It plays a key role in many fields, such as Pattern recognition, Magnetic resonance data analysis, time series analysis, Windows error processing, Image process-ing。 Model selection is a key part of statistical methods, and variable selection is one of its crucial problems. This paper proposes a new criterion based on maximal information coefficient(MIC) of model selection and variable selection, and there is an application of MIC to Magnetic resonance data analysis.1. Model selection based on MIC· A procedure of MIC written in C++language is given, which is much more flexible than the one given by David.N.Reshef et al;· A new method in model selection-minimizing the association between residue and explanatory variable is proposed, which is robust to different types of noise, and also an evaluation standard of different model selection criterion;2. Variable selection based on MIC An extension of MIC to Partial MIC is given, which can be used to· Find the inhibited association (Linearity or nonlinearity);· Find the true model;· Make full use of the data;· Find the reason of the dependency;3. Magentic resonance data analysis (Violence tendency in teenagers) We find that· The number of edge within Occipital lobe of patients is less than that of normal ones;· The function of Limibic lobe of patients is weaker than that of normal ones;· The number of edge between Central region and Parietal lobe, Frontal lobe and Lim-ibic lobe, Parietal lobe and Occipital lobe of patients is less than that of normal ones;· The number of edge between Frontal lobe and Temproal lobe of patients is more than that of normal ones;...