IMPROVED VERSION OF DIFFERENTIAL EVOLUTION FOR UNSUPERVISED FEATURE SELECTION

Among the many facets of machine learning, pattern recognition methods have been extensively used to address a broad range of practical issues. Given the current climate of huge datasets, many traits are likely to be redundant or associated with one another. To improve data accuracy and purity, it seems sense to eliminate such superfluous, noisy aspects. Reducing the number of features in a model not only speeds up subsequent data processing but also improves the model's capacity to generalize. To better select features without human oversight, this work examines a revised form of differential evolution.