A Coevolutionary approach for classification problems: Preliminary results

In the current classification problems, how to eliminate redundant features, find out important features, and choose appropriate classifiers for these feature set play a vital role. This paper presents a co-operative co-evolution approach (COCEA) with dual populations for optimizing both the artificial neural network (ANN) model and feature subset selection simultaneously. In COCEA, each feature subset is encoded as a binary string. Meanwhile, an ANN is represented in a matrix-form with real values of its weight and bias. During the process of evolution, a co-Operation mechanism is used to integrate the two populations and the final solution is the combination of the two most elite individuals of each population in a hope that the final solution will satisfy both ANN optimization and feature subset selection. The performance of COCEA is examined on both the well-known benchmark problems and Oil Spill dataset in SAR Images. In comparison with the other algorithms, experimental results illustrate that COCEA can significantly outperform other peer algorithms in terms of classification accuracy. © 2018 IEEE.