Regression Techniques in Software Effort Estimation Using COCOMO Dataset

Regression techniques are used to measure software estimates accuracy for evaluation and validation. The common evaluation criteria in software engineering like Magnitude Relative Error (MRE) that computes absolute error percentage between actual and predicted efforts for reference samples is used. The Mean Magnitude Relative Error (MMRE) and Median Magnitude Relative Error (MdMRE) are the de facto standard evaluation criterion to assess the accuracy of software prediction models. The regression algorithms like M5 algorithm and Linear Regression in Software Effort Estimation using COCOMO dataset is evaluated. Simulation results demonstrate that the errors such as MMRE and MdMRE of M5 algorithm is less than linear regression in forecasting by 80.20 and 45.30 percentage respectively. Future work aims to reduce further the error of forecasting.