Extreme Learning Machine Applied To Software Development Effort Estimation

Abstract

Reliable estimation of software development effort is essential for effective project planning, scheduling, and resource allocation. Conventional estimation approaches such as expert judgment and algorithm-based models are often affected by human bias and rigid assumptions. This paper presents a machine-learning driven effort estimation framework based on the Extreme Learning Machine (ELM). The proposed framework uses historical project data from the COCOMO-81 repository and models the relationship between project attributes and actual development effort. ELM is implemented and compared with Linear Regression, K-Nearest Neighbour, Support Vector Machine, and Multilayer Perceptron models. Standard error-based performance indicators, including MAE, MSE, RMSE, and MMRE, are employed for quantitative evaluation. In addition, statistical significance of results is verified using Shapiro–Wilk and Wilcoxon signed-rank tests. Experimental observations confirm that the ELM model delivers superior predictive accuracy with significantly lower training time, demonstrating its suitability for practical software project management environments.