An Improved Feature Selection using Maximized Signal to Noise Ratio Technique for TC

Aim of this work is to produce excellent accuracy with reduced feature set by a simple method. When the profile built using a feature selection method called MSNR (maximized signal to noise ratio) combined with modified fractional similarity method, it performs in a competitive manner. MSNR identifies the highly contributing features and increases the distance between the profiles. Experimental results show that when we select only top 3% features of each class using MSNR (maximized signal to noise ratio) and use these profiles in combination with modified fractional method, achieved 90% classification accuracy