The performance of differential evolution algorithm for training CSFNN using a pattern recognition application

In this work, Conic Section Function Neural Network (CSFNN) has been trained by differential evolution algorithm (DEA) to overcome local minimum problems. The classification performance of the CSFNN trained by DEA has been analyzed by using high-dimensional and non-linear signature recognition database. The CSFNN training performance of the DEA has been compared with that of the gradient based back-propagation algorithm (BPA). The simulation results show that the classification performance of the CSFNN trained by DEA is more stable than that of the CSFNN trained by BPA for running several trials.