Enhancing Software Effort Estimation in Agile Projects Through Expert-Guided Feature Selection and Machine Learning

Abstract

This study proposes a hybrid methodology that integrates expert judgment and machine learning techniques to improve software effort estimation in Agile projects. A comprehensive survey of 45 software professionals identified key influencing attributes, which were further refined using recursive feature elimination (RFE). The performance of three machine learning models—Linear Regression, Multilayer Perceptron (MLP), and Support Vector Regression (SVR)—was evaluated. Results showed that incorporating expert-informed features significantly improved the accuracy of predictions, with Linear Regression achieving the best results (R² = 0.9319, MAE = 1726).