SEET:: Software Development Effort Estimation Using Ensemble Techniques

Software development effort estimation (SDEE) is a significant activity in project management and serves as the basis for project bidding, planning, staffing, resource allocation, scheduling, and cost estimation. The accuracy of SDEE techniques varies from project to project, which makes them rather unreliable. In this backdrop, we propose a foundation centered ensemble-based SDEE approach. The primary goal of this approach is to design an ensemble consisting of different machine learning methods for improving the prediction accuracy of SDEE. In recent times, several research results have been reported on machine learning based ensemble design, but extreme learning machine (ELM) and least square support vector regression (LSSVR) have not been used to develop an ensemble. We chose three machine learning techniques, namely ELM, LSSVR, and multilayer perceptron (MLP) as the base techniques to build an ensemble. We investigated the performance of a homogeneous ensemble design using a linear combination rule with standardized accuracy as a weight factor. The performance of the ensemble model is validated and compared with root mean square error (RMSE) based weighted average ensemble model with equivalent configuration. The experimental study was conducted using publicly available PROMISE repository test suite. We achieved promising results for SEET model compared to base learners and RMSE ensemble model.