Improving Object Point Software Effort Estimation Using a Deep Learning Approach

Abstract

Delivering high-quality software within budget and schedule constraints is a critical responsibility of project managers, and it depends on the accurate early prediction of software development effort. However, previous software projects' effort estimation methods, such as source lines of code and function point size metrics, have limitations in estimating effort at the specification stage. A machine learning-based object point analysis technique for software effort estimation is the earliest technique to predict software project effort at the specification stage. Even object point analysis software and effort estimation techniques using machine learning algorithms can make effort predictions at an earlier stage. Still, they have some accuracy limitations compared to the actual effort. However, deep learning models showed better accuracy in effort estimation near the actual effort at the earliest specification stage. This work advances the state-of-the-art by combining OPA’s simplicity with DNN’s predictive power, offering a practical solution for industry adoption. The study conducted a comparative experimental study on object point effort estimation using a recent deep learning model and three traditional machine learning algorithms (K-Neighbors Regression, Linear Regression, and Support Vector Regression). The experimental results ensured that the Deep learning model achieves superior performance, with minimal prediction errors (lower MAE, RMSE, and MSE) and a higher R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score compared to traditional approaches.