Prediction of Software Cost Estimation Using Stacking Ensemble Learning method

The process of accurate cost estimation is one of the important factors for the success of software projects, so it is necessary to understand the process of estimating the cost and planning in advance to manage the project positively and effectively, which in turn helps the project manager and the client to determine the work budget. This paper introduced Ensemble learning method of stacking. This method works in two stages: the first is to train many individual models called models with basic learners, and the second is to collect model expectations from the first stage as an introduction to an intermediate model called the meta-model that works on the final prediction. The following models (Random Forest (RF), Linear Regression (LR), AdaBoost, XGBoost, Gradient Boost, and K_Nearest Neighbor (K_NN)) are basic learners for the stacking method. The learners were trained on the ISBSG dataset from the ISBSG repository. At the beginning of the work, the dataset was processed to obtain high prediction accuracy, and the characteristics that suit the cost estimation process were determined. The proposed method was applied to it, and the model performance was evaluated using the following accuracy measures (MAE, RMSE, R-squared, MMRE). The results of implementing the stacking model gave a prediction accuracy of 98% and a lower error rate compared to the individual models by 0.0927, thus proving its superiority over the individual models by increasing the estimation accuracy. The results of implementing the algorithm for the accuracy measures, respectively (207,533,0.983,0.0927). The proposed method contributes to producing a software development cost estimation tool with an accuracy of 98%, which in turn helps estimators in pre-estimating the cost and effort with high accuracy, which leads to the success of the work.