Improving the accuracy in software effort estimation: Using artificial neural network model based on particle swarm optimization

Recent years, the software industry is growing rapidly and people pay more attention on how to keep high efficiency in the process of software development and management. In the process of software development, time, cost, manpower are all critical factors. At the stage of software project planning, project managers will evaluate these parameters to get an efficient software develop process. Software effort evaluate is an important aspect which includes amount of cost, schedule, and manpower requirement. Hence evaluate the software effort at the early phase will improve the efficiency of the software develop process, and increase the successful rate of software development. This paper proposes an artificial neural network (ANN) prediction model that incorporates with Constructive Cost Model (COCOMO) which is improved by applying particle swarm optimization (PSO), PSO-ANN-COCOMO II, to provide a method which can estimate the software develop effort accurately. The modified model increases the convergence speed of artificial neural network and solves the problem of artificial neural network's learning ability that has a high dependency of the network initial weights. This model improves the learning ability of the original model and keeps the advantages of COCOMO model. Using two data sets (COCOMO I and NASA93) to verify the modified model, the result comes out that PSO-ANN-COCOMO II has an improvement of 3.27% in software effort estimation accuracy than the original artificial neural network Constructive Cost Model (ANN-COCOMO II).