Feature Selection Based on Discriminant and Redundancy Analysis Applied to Seizure Detection in Newborn

The role of feature selection is fundamental in pattern recognition, increasing accuracy and lowering complexity and computational cost in the presence of redundant and irrelevant features. This paper outlines a new feature selection algorithm based on discriminant and redundancy analysis to determine the "goodness" of feature subsets. The performance of this method was compared to a correlation-based feature selection method via relevance and redundancy analysis. To evaluate their effectiveness for seizure detection in newborn, the features extracted from seizure and non-seizure segments were ranked by these methods. Then, the optimized ranked feature subsets were fed to multilayer backpropagation neural networks as the classifiers. The classifier performance was used as indicator of the feature selection effectiveness. The results showed an average seizure detection rate of 90%, an average non-seizure detection rate of 91%, an average false rejection rate of 91% and an average detection rate of 90%. Our feature selection method allows a feature reduction up to 80%