Software effort estimation using functional link neural networks tuned with active learning and optimized with particle swarm optimization

Abstract

This paper puts forward a new learning model based on the collaborative effort of active learning and particle swarm optimization (PSO) in functional link artificial neural networks (FLANNs) to estimate software effort. The active learning uses quick algorithm to detect the essential content of the datasets by which the dataset is reduced and are processed through PSO optimized FLANN. The PSO uses the inertia weight, which is an important parameter in PSO that significantly affects the convergence and exploration-exploitation in the search space while training FLANN. The Chebyshev polynomial has been used for mapping the original feature space from lower to higher dimensional functional space. The method has been evaluated exhaustively on different test suits of PROMISE repository to study the performance. The computational results show that the active learning along with PSO optimized FLANN greatly improves the performance of the model and its variants for software development effort estimation.