Unsupervised similarity-based feature selection using heuristic Hopfield neural networks

An unsupervised similarity-based feature selection approach using heuristic Hopfield neural networks (UFS-HHNN) is presented. The key novel ingredient of the algorithm is to formulate the feature selection problem as a combinatorial optimization problem. To our best of knowledge, this is the first attempt at formulating feature selection as a combinatorial optimization problem. We map the feature selection problem to a single layered Hopfield Networks and adjust parameters. Maximum Information Compression Index (MICI), the amount of reconstruction error committed if the data is projected to a reduced dimension in the best possible way, is employed as a similarity measure. Simulation on eight benchmark datasets with different dimensions and size shows that feature subsets with much lower redundancy are achieved by UFS/spl I.bar/HHNN than the recently developed unsupervised algorithm. Our approach can be easily extended to supervised feature selection and feature scaling.