Software Enhancement Effort Estimation using Machine Learning Regression Methods

Abstract

Software enhancement must be carefully planned and taken to satisfy customer change requests, such as adding a new functionality and deleting or changing an existing one. A poorly constructed planning may cause project failures to meet budget targets and deadlines. One of the software project planning activities is effort estimation. In this paper, we investigate the effectiveness and performance of four Machine Learning Regression Methods (MLRM): Ada Boost Regressor (ABR), Gradient Boosting Regressor (GBR), LinearSupport Vector Regression (LinearSVR), and Random Forest Regression (RFR) to predict software requirements enhancement effort. The analysis was based on the results of experiments carried out on real projects in the software industry. These techniques were trained and tested with six software development project datasets including functional requests and the PROMISE repository including enhancement requests. The results of enhancement effort with different machine learning techniques were compared with the enhancement effort obtained from the expert judgement. The best performances were observed with RFR in terms of: MAE (Mean Absolute Error) = 0.040, mean square error (MSE)= 0.045 and root mean square error (RMSE)= 0.215. Therefore, RFR could be recommended for the estimation of software enhancement effort when using expert judgment.