AN APPROACH TO ESTIMATE THE DURATION OF SOFTWARE PROJECT THROUGH MACHINE LEARNING TECHNIQUES

Anam Khalid; Muhammad Ahsan Latif; Muhammad Adnan

Anam Khalid Department of Computer Science, University of Agriculture, Faisalabad, Pakistan
Muhammad Ahsan Latif Department of Computer Science, University of Agriculture, Faisalabad, Pakistan
Muhammad Adnan IMIS, Department of computer science & information engineering, NCKU, Taiwan

Keywords: Duration of Software Project, Artificial Neural Network, Feed Forward Neural Network, Bayesian regularization back propagation

Abstract

In the software project, to estimate the duration of software processes is frequently a complex problem. Only 39 percent projects are finished on time relative to the original schedule. Many research efforts had been developed to estimate the duration, but no single model could be used which was suitable for this problem. It is a challenging task to recognize a reliable model for estimation. Due to wrong selection for model or assigning weight, a software system faced many problems which lead to cost, time, effort and schedule overrun. This research proposed a procedure to estimate the duration of software projects by applying machine learning technique. The Bayesian regularization back propagation (BR) and Levenberg–Marquardt (LM) training algorithms are used within Feed forward neural network and Radial base neural network and got results of both models. This approach is applied to the data which is taken from the literature review. After training of the models consuming both training algorithms, it is concluded that BR offers superior results than LM.