Estimation model for enhanced predictive object point metric in OO software size estimation using deep learning

Abstract

The Software industry’s rapid growth contributes to the need for new technologies. PRICE software system uses Predictive Object Point (POP) as a size measure to estimate Effort and cost. A refined POP metric value for object-oriented software written in Java can be calculated using the Automated POP Analysis tool. This research used 25 open-source Java projects. The refined POP metric improves the drawbacks of the PRICE system and gives a more accurate size measure of software. This paper uses refined POP metrics with curve-fitting neural networks and multi-layer perceptron neural network-based deep learning to estimate the software development effort. Results show that this approach gives an effort estimate closer to the actual Effort obtained through Constructive Cost Estimation Model (COCOMO) estimation models and thus validates refined POP as a better size measure of object-oriented software than POP. Therefore we consider the MLP approach to help construct the metric for the scale of the Object-Oriented (OO) model system.