Feature selection for binary classification based on class labeling, SOM, and hierarchical clustering

Feature selection plays an important role in algorithms for processing high-dimensional data. Traditional pattern classification and information theory methods are widely applied to feature methods. However, traditional such as Fisher Score, Laplacian relief use class labels inadequately. Previous based MIFS ignore the intra-class tight inter-class sparse property of samples. To address these problems, a algorithm binary problem is proposed, which on label transformation using self-organizing mapping neural network (SOM) cohesive hierarchical clustering. The first converts without numerical meaning into values that can participate operations retain through mapping, constitutes two-dimensional vector from it attribute be judged. Then, vectors clustered by SOM Finally, evaluation function value calculated, closely related intra-cluster tightness, inter-cluster separation, division accuracy after clustering, used evaluate ability alternative attributes distinguish between classes. It experimentally verified robust effectively screen with strong improve prediction performance classifier.