Harmony and bio inspired harmony search optimization algorithms for feature selection in classification

摘要

Today, constant and rapid changes in information and communication technologies offer unanimous access to vast amounts of information and make an exponential increase of the amount of data available online. High dimensionality has always been a great challenge for all learning algorithm and "curse of dimensionality" has been studied for a long time. The high-dimensional feature vectors often impose a high computational cost when classification is performed. Feature selection plays major role as a pre-processing technique in reducing the dimensionality of the datasets in the fields of data analysis and data mining applications. This paper presents two music inspired novel Harmony Search optimization algorithms, HS-1-NN and Bio-HS-1-NN, for wrapper feature selection where Bio-HS-1-NN is augmented with food source exploitation behaviour of honey bees, which improves the members of the Harmony memory based on their fitness values and hence speeds up the convergence speed of the search. 1-NN classifier has been used to evaluate the quality of the solutions. The performance of the proposed approaches has been analysed by experiments with various real-world data sets and the proposed approaches, exhibited better performance than other methods in terms of classification accuracy and convergence rate.