Software effort estimation using cascade neural network optimised based on modified particle swarm optimisation (MPSO-CNN)

Software effort estimation has a significant role in software development engineering. The inaccurate estimation will increase the failure possibilities of the project. On the contrary, accurate estimation enables the project developers to finalise the projects within the required time and budget. Furthermore, it is considered a big challenge to obtain the satisfactory accuracy of project development at the beginning. To tackle this problem, soft computing techniques such as artificial neural network (ANN) has already demonstrated a remarkable performance in software effort estimation. However, the optimal weights for the neural network are still considered a big dilemma. In this paper, a cascade neural network (CNN) is optimised based on modified particle swarm optimisation (PSO). The modified PSO can overcome the premature convergence of PSO as well as avoid falling into local optima effectively. The experimental results have shown the superiority of the proposed work compared with the standard PSO significantly.