Empirical Evaluation of Different Machine Learning Methods for Software Services Development Effort Estimation Through Correlation Analysis

The concept of development effort generally means the time or the cost of developing a software service. An essential factor to successfully manage and control a project is the accurate estimation of the development effort and an over and underestimation lead to the loss of project resources. So far, different effort estimation models have been presented in three domains: expert judgment, algorithmic methods and machine learning methods. Recently, several approaches in the last domain, machine learning, have been applied for software service development effort, which had a higher performance in comparison to the other two domains. This paper presents an empirical evaluation of the performance and accuracy of five main machine learning methods using the correlation analysis approach and investigates the effects of feature selection on the estimation accuracy. The evaluations and comparisons are performed using two well-known and real-world datasets, NASA and ISBSG and two artificial datasets, Moderate and Severe. Finally, the obtained results provide a clear illustration of the performance of these machine learning methods and the effects of feature selection on the estimation accuracy.