Implementation of Prediction Model for Object Oriented Software Development Effort Estimation using One Hidden Layer Neural Network

The prediction model for object-oriented software development effort estimation using one hidden layer neural network has been implemented in this paper. This prediction model has been empirically validated on PROMISE software engineering repository dataset. Accurate prediction of software development effort and schedule is still a challenging job in software industry. This prediction model has been implemented through programming in MATLAB using one hidden layer feed forward neural network(OHFNN) and results obtained from this program are compared with existing algorithms like traingda and traingdm of NNTool. By a large number of simulation work OHFNN 16-19-1 is found optimal structure for this prediction model. OHFNN 16-19-1 means 16 neurons in input layer, 19 neurons in hidden layer and 1 in output layer. Training of the neural network has been done by using back propagation with a gradient descent method. Performance of predictor is better in terms of accuracy than existing well established constructive cost estimation model (COCOMO). In this network, convergence is obtained by minimizing the root mean square error of the input patterns and optimal weight vector is determined to predict the software development effort.