Home Publications Automated Selection of a Software Effort Estimation Model Based on Accuracy and Uncertainty

Automated Selection of a Software Effort Estimation Model Based on Accuracy and Uncertainty

Fatih Nayebi,Alain Abran,Jean-Marc Desharnais
August 15, 2015
Artificial Intelligence Research

Summary

This paper addresses the critical challenge of selecting the most appropriate software effort estimation model for different project contexts. The research introduces a novel automated approach that considers both estimation accuracy and uncertainty when evaluating and selecting among competing estimation models. The proposed method helps project managers and software teams make more informed decisions about which estimation techniques to apply, based on quantifiable metrics rather than subjective preferences or organizational habits.

Key Insights

  1. Different software effort estimation models perform better or worse depending on project characteristics, making model selection a context-dependent decision
  2. Considering only accuracy metrics (like MMRE or MdMRE) provides an incomplete picture of estimation model performance
  3. The integration of uncertainty analysis with traditional accuracy metrics offers a more robust framework for model selection
  4. Automated selection mechanisms can adapt to specific organizational contexts and historical project data
  5. The research demonstrates a 15-28% improvement in estimation accuracy when using the proposed automated selection approach compared to consistently using any single model

Impact

This research has contributed to more realistic and reliable software project planning by offering organizations a systematic way to select estimation models. By introducing uncertainty as a critical factor in the selection process, the work has helped advance the state of the art in software effort estimation beyond simple accuracy metrics. The automated selection approach has particular value for organizations with diverse project portfolios, where no single estimation model consistently outperforms others across all project types.

Access

The full paper is available at Sciedu Press - Artificial Intelligence Research.

Resources