Cost sensitive discriminant analysis by swarm intelligence for access control system

Misclassification, which is a common phenomenon due to the absence of a perfect classifier, incurs losses from different classes. This paper aims on developing an algorithm which incorporates a cost sensitive subspace reduction technique for extracting the cost sensitive features from facial images of test subjects. The weight vector which projects the faces on to a new domain with lesser dimensions is obtained by globally optimizing a cost function using swarm intelligence. The weight vector consisting of the cost sensitive features has been obtained by choosing suitable cost values for different classes so that the error obtained can be mostly attributed to false rejection rather than false acceptance. The images with reduced features are divided into training and testing sets and fed to a neural network classifier. The accuracy obtained has been found to be higher than many state of the art classification techniques. Running the proposed technique has further led to affirmation of the primary motive which is keeping false acceptance as low as possible.