Effective Software Effort Estimation using Heterogenous Stacked Ensemble

Software Effort Estimation (SEE) estimates the effort required to complete the project successfully. The accuracy of estimation is directly proportional to the project success. This paper proposes a heterogenous stacked ensemble for effective effort estimation with Artificial Neural Network (ANN) and Support Vector Regressor (SVR) as base learners. Further, an empirical comparison is made among the proposed model and base learners to figure out the accuracy of proposed model. Five datasets from PROMISE repository are used and accuracy measures for comparison are MAR (Mean Absolute Error) and MMRE (Mean Magnitude of Relative Error). It is found that proposed model improves the performance by reducing the MAR by 50.4% and MMRE by 54.6% of base models respectively. It can be concluded from the experimental results that the proposed heterogenous stacked ensemble is best performer among the candidate models for SEE statistically.