Improved Feature Subset Selection using Hybrid Ant Colony and Perceptron Network

As classification accuracy is strongly dependent on the set of features used as input variables. For automatic feature acquisition, the literature provides us with numerous strategies aiming to find a “best” set of features. Thus we need to find a “best” set of features given a constraint on the computational complexity or cost of the feature acquisition, which may dominate the cost of the classifier. In our work we improve on the accuracy of ACO FSS algorithm by incorporating the Perceptron Classifier as fitness function of ACO whose classification error is taken as the Cost of Classifier. This method improves on accuracy of selecting minimum number of features with maximum accuracy achieved.