The Ensemble of Ant Colony Optimization and Gradient Descent Technique for Efficient Feature Selection and Data Classification

Abstract

Feature selection is of the paramount importance in the domain of data classification task especially when the number of features is huge. Further, gradient descent is an important technique for training the perceptron network and to find the hyper-plane that can classify the nonlinearly classifiable data sets with least error. However, modifying the synaptic weights of each input neuron is cumbersome and also time taking. In this work, an ensemble approach of ant colony optimization (ACO) and gradient descent technique is used. In this ensemble model, ACO will first select the important features, and subsequently, the gradient descent technique will modify the synaptic weights of these input features. This combined approach renders reduced training time of the perceptron network with enhanced classification accuracy of the test data sets. Experimental analysis and validation of proposed ensemble technique has also been incorporated for standard data sets. Over and above, the proposed technique have also been implemented for the purpose of identifying different commonly used mechanical parts.