Artificial Neural Network-Based Software Effort Estimation with Project Metrics

Abstract

Software effort estimation requires reliable use of historical project evidence because inaccurate effort forecasts can affect staffing, scheduling, budgeting, and delivery commitments. This article presents an Artificial Neural Networkbased effort estimation framework that learns from historical project metrics such as project size, planned duration, team size, developer experience, requirement volatility, reuse percentage, technical complexity, testing effort, defect count, and application domain. The framework prepares completed project records through data cleaning, missing-value review, normalization, categorical encoding, training-validation-test partitioning, and repeated ANN training to reduce unstable estimation behavior. The study shows that ANNbased estimation is more reliable when project categories are analyzed separately, because small, medium, large, high-volatility, and reuse-intensive projects show different error patterns. The proposed approach supports project planning by providing a data-driven effort baseline while still requiring managerial adjustment for unusual projects, unstable requirements, new technologies, and high integration complexity.