Abstract:Today, modern databases with "Big Dimensionality" are experiencing a growing trend. Existing approaches that require the calculations of pairwise feature correlations in their algorithmic designs have scored miserably on such databases, since computing the full correlation matrix (i.e., square of dimensionality in size) is computationally very intensive (i.e., million features would translate to trillion correlations). This poses a notable challenge that has received much lesser attention in the field of machine learning and data mining research. Thus, this paper presents a study to fill in this gap. Our findings on several established databases with big dimensionality across a wide spectrum of domains have indicated that an extremely small portion of the feature pairs contributes significantly to the underlying interactions and there exists feature groups that are highly correlated. Inspired by the intriguing observations, we introduce a novel learning approach that exploits the presence of sparse correlations for the efficient identifications of informative and correlated feature groups from big dimensional data that translates to a reduction in complexity from O(m2n) to O(mlogm + Ka mn), where Ka << min(m,n) generally holds. In particular, our proposed approach considers an explicit incorporation of linear and nonlinear correlation measures as constraints in the learning model. An efficient embedded feature selection strategy, designed to filter out the large number of non-contributing correlations that could otherwise confuse the classifier while identifying the correlated and informative feature groups, forms one of the highlights of our approach. We also demonstrated the proposed method on one-class learning, where notable speedup can be observed when solving one-class problem on big dimensional data. Further, to identify robust informative features with minimal sampling bias, our feature selection strategy embeds the V-fold cross validation in the learning model, so as to seek for features that exhibit stable or consistent performance accuracy on multiple data folds. Extensive empirical studies on both synthetic and several real-world datasets comprising up to 30 million dimensions are subsequently conducted to assess and showcase the efficacy of the proposed approach.

Mining Non-Redundant High Order Correlations in Binary Data

Mining Associated and Item-Item Correlated Frequent Patterns

An Improved Algorithm for Mining Non-Redundant Interacting Feature Subsets

Feature Selection with Conditional Mutual Information Considering Feature Interaction

Mining Top-k Minimal Redundancy Frequent Patterns over Uncertain Databases.

A new dependency and correlation analysis for features

Finding High-Order Correlations In High-Dimensional Biological Data

An algorithm for mining strongly correlated pairs in relational table

R2CI: Information theoretic-guided feature selection with multiple correlations

A Set of Efficient Methods to Generate High-Dimensional Binary Data with Specified Correlation Structures

Making Trillion Correlations Feasible in Feature Grouping and Selection

Feature selection based on mutual information with correlation coefficient

Feature Selection Method Based on Differential Correlation Information Entropy

Mining maximal correlated member clusters in high dimensional database

Efficiently mining mutually and positively correlated patterns

TCIC_FS: Total Correlation Information Coefficient-Based Feature Selection Method for High-Dimensional Data

Information-theoretic Feature Selection Based on the Weight of the New Classification Information

Redundant feature selection by conditional linear correlation

A unified framework for correlation mining in ultra-high dimension

A Nonlinear Correlation Measure for Multivariable Data Set

Feature selection based on network maximal correlation