Improving software cost estimation process using feature selection technique

Software cost estimation is an important task in any software development cycle. It helps project and software engineers to do better planning and resource management. Mining historical data to predict the future cost is a commonly used approach. However, dataset used in software estimation models plays a crucial role in the model accuracy. In this paper, feature selection techniques applied to different datasets in order to improve the model accuracy, reduce data redundancy, and increase the model performance. Five types of evaluation criteria were used to compare the results of 13 machine-learning methods before and after applying the feature selection techniques. Results showed that feature selection techniques can obviously increase the accuracy of the prediction model.