Improving Effort Estimation Accuracy in Software Development Projects Using Multiple Imputation Techniques for Missing Data Handling
Shahida Hayat, Wajahat Akbar, Tariq Hussain, Muhammad Inam Ul Haq, Altaf Hussian, Irshad Khalil, Muhammad Nawaz Khan, Samsonova Diana · ICCK Transactions on Intelligent Systematics · 2024
Intelligent project management systems rely on high-quality historical data for accurate automated decision-making, yet missing data in software project repositories remains a persistent challenge that degrades intelligent estimation performance. This study proposes an Intelligent Decision Support Framework (IDSF) for software development effort estimation (SDEE) that integrates Multiple Imputation (MI) as a critical data quality enhancement layer within the Analogy-Based Effort Estimation (ABEE) model. The framework is evaluated on the ISBSG dataset by systematically comparing six imputation strategies. Results demonstrate that the MI-enhanced framework achieves competitive and more stable MMRE values while fully preserving dataset integrity, in contrast to traditional deletion methods that cause substantial data loss. Additionally, a theoretical analysis of Long Short-Term Memory (LSTM) networks is provided as a prospective deep learning estimator, highlighting that high-quality restored data is structurally necessary for effective LSTM training. This work contributes to intelligent systems in software engineering by establishing MI as a robust data quality module, laying a strong foundation for building more reliable AI-driven intelligent project management systems and advancing intelligent systematics research.