10.1007/978-3-032-22211-4_2">
 

Cost Estimation of DoD ACAT 1 Software Programs: Statistical Regression vs. Neural Networks

Document Type

Conference Proceeding

Publication Date

6-14-2026

Abstract

Accurately estimating software costs is crucial for Department of Defense (DoD) projects to avoid budget overruns and resource misallocation. This study compares the effectiveness of statistical regression techniques and neural network models for cost estimation, using 306 records with ESLOC, SLOC, and project attributes. A baseline model had an MAE of 1.35 and MSE of 2.82. The refined Ordinary Least Squares (OLS) regression model achieved an R2 of 0.58, Adjusted R2 of 0.57, PRESS R2 of 0.57, MAE of 0.85, and MSE of 1.20, focusing on key predictors such as ESLOC and development hours. In contrast, the best-performing neural network achieved an R2 of 0.54, MAE of 0.88, and MSE of 1.27, with L2 regularization reducing overfitting but lowering R2 to 0.48. The findings highlight the strengths of OLS regression in small-sample scenarios, where its transparency and reliability make it better suited to DoD cost analysis than neural networks, which often require extensive tuning and face instability. These results underscore the value of statistical methods in providing actionable insights for refining cost estimation frameworks and guiding resource management in Agile and traditional DoD environments.

Comments

© The Authors, under exclusive license to Springer Nature Switzerland AG 2026.

The full paper is available from the publisher via subscription or purchase using the DOI link below.

Event: 23rd International Conference, CSC 2025, and 21st International Conference, FCS 2025, Held as Part of the World Congress in Computer Science, Computer Engineering, and Applied Computing, CSCE 2025, Las Vegas, July 21-24, 2025.

Source Publication

Emerging Trends in Scientific Computing and Theoretical Computer Science

This document is currently not available here.

Share

COinS