Classification is a data mining approach employed for predicting the group membership for data instances. Various classification methods exists, which can be utilized for classification. In the existing work earlier, Neural Network incorporated with Intuitive, Interpretable Correlated-Contours fuzzy rules (IC-FNN) aimed at function approximation was introduced, which is helpful in obtaining correlated fuzzy rules and non-separable fuzzy rules coming under the right optimization problem. Also, Hybridization of Ant Colony Optimization Genetic Algorithm (HACOGA) was suggested for rectifying these problems. It fine-tunes the parameters of the fuzzy rules extracted. Hybridization is used on specific factors and ACO and GA variables, which share few features in the computation. However, the available fuzzy neural network is affected by problems associated with the number of neurons. Generally, neurons make use of activation functions popularly found to get the network response finally. In order to get over these problems in this research work, density based regularization methods and activation functions are presented for the neural network model, permitting the less important neurons to be eliminated. Also the parameter of the fuzzy rules is refined with the aid of Hybrid Ant Colony Particle Swarm Optimization (HASO) to minimize the complexity and search space. The results obtained from experiments reveal that proposed (HASO) help in the performance regularization of FNN in terms of recall, precision, accuracy and F-measure for the Abalone age prediction dataset.
Volume 12 | 03-Special Issue