International Journal of Computational Intelligence Research ISSN 0973-1873 Volume 7, Number 3 (2011), pp. 271-294 © Research India Publications http://www.ripublication.com/ijcir.htm On Genetic Algorithm and Multiple Preprocessors Assisted Feature Boosting for Electronic Nose Signal Processing Prabha Verma 1 , Divya Somvanshi 2 and R.D.S. Yadava 3 Sensors & Signal Processing Laboratory, Department of Physics, Faculty of Science, Banaras Hindu University, Varanasi 221005, India. E-mail: 1 pverma.bhu@gmail.com, 2 somvanshi.divya@gmail.com, 3 ardius@gmail.com Abstract The paper presents a method for feature extraction that explores data space through different preprocessing strategies in combination with principal component analysis (PCA) and genetic algorithm (GA). A preprocessor/PCA combination transforms data space into feature space; a change in preprocessor results in an alternate feature space. The proposed method first fuses the feature spaces by simple concatenation of the alternate feature vectors then allows them to evolve genetically. The genetic evolution of each fused feature vector is based on treating the feature vector as chromosome and the feature components as genes. The initial population is created on the basis of a probability distance metric. The fitness and ranking is done by using PCA generated variances as measure of significance. In the terminal population the frequency of a gene (principal component) occurrence is interpreted as a measure of its significance in defining the feature vector. Finally, the feature components are given additional weight according to ( ) j j ij ij p p z z 2 log 1 − = where ij z denotes j-th feature component of i-th sample in fused feature space and j p denotes the probability of j-th component occurring in the terminal population. In order to demonstrate the efficacy of this idea we employed only two well known preprocessing methods: vector autoscaling and dimensional autoscaling. The feature vectors defined in this manner were used as input a radial basis neural network classifier for validation. Several benchmark datasets (both chemical and non-chemical) available from open sources were used in the analysis for validation. It is found that the scheme of feature fusion and weighting enhances classification rate in most cases analyzed. Keywords: Feature extraction, electronic nose, feature fusion, genetic algorithm, pattern recognition.