Software Cost Estimation Using LSTM-RNN

Abstract

Any project, however, astronomically immense or minute, or regardless of the industry that is undertaken, needs to be performed and distributed under certain constraints. Cost is one of those constraints that project management needs to efficaciously control. Cost may be the driving force or an obstacle in deciding project’s future. Consequently, a trustable and meticulous effort estimation model is a perpetual challenge for project managers and software engineers. Long short term memory (LSTM) and recurrent neural networks (RNN) are utilized in this paper to suggest the incipient model for calculating the effort needed to develop a software. The result of the model is evaluated on COCOMO’81, NASA63 and MAXWELL datasets. The experimental results showed that LSTM-RNN with linear activation function (LAF) amends the precision of cost estimation in comparison with other models utilized in the study.