Risk integrated effort estimation of software projects: a comparative analysis of machine learning techniques

Accurate software project effort estimation and risk management are the pillars of delivering an on-time, within budget and quality project. In our earlier research, a formula for computing risk integrated effort estimate by adding weighted cost of risk management for each project cost factor to the cost of initial effort estimate of the project has been proposed. In this research, neural network techniques: multilayer perceptron (MLP), general regression neural network (GRNN), cascade correlation neural network (CCNN) and radial bias function neural network (RBFNN); support vector regression (SVR), and adaptive neuro fuzzy inference system (ANFIS) to obtain the integrated effort estimate as close as possible to the actual effort spent on the project have been applied. The techniques have been tested on two datasets: Agile and Waterfall datasets. GRNN gave the best results in terms of lowest values of accuracy measures: mean absolute error (MAE), mean magnitude of relative error (MMRE), mean balanced relative error (MBRE), and mean inverted balanced relative error (MIBRE). This research also compares performance of GRNN with evolutionary algorithms artificial bee colony (ABC), particle swarm optimisation (PSO) and global local binary particle swarm optimisation (GLBPSO), and the results for GRNN are comparable.