Date of Award

9-1996

Document Type

Thesis

Degree Name

Master of Science

First Advisor

Daniel Ferens, PhD

Abstract

The rising number and importance of Department of Defense software developments, when combined with declining defense budgets, has resulted in a critical need to accurately plan and manage software development costs and schedules. Unfortunately, the increasing size, complexity, and diversity of these software developments has made accurate estimating problematic. Uncalibrated software cost models have not generally produced reliable results due to generic default parameters and improper usage. The default parameters cannot hope to accurately represent and predict the wide variability of software efforts to which the models are being applied. However, some of the models have achieved improved accuracy by calibration from their generic default parameters to new parameter values based on specific characteristics of the development efforts being estimated. This research effort focused on the calibration of SoftCost-R, Version 8.4, to specific stratified data sets obtained from the Space and Missile Systems Center (SMC) Software Database, Version 2.1, (SWDB). The accuracy of the new calibrated inputs was verified through comparisons between the calibrated and default estimates and the actual cost data. Statistical methods used to make the comparisons included magnitude of relative error (MRE), mean magnitude of relative error (MMRE), root mean square (RMS), relative root mean square (RRMS), and prediction level Pred (k/n) or percentage of estimates within (100 * k/n)% of the actual costs. The new calibrated parameters resulted in more accurate effort estimates and the calibration method appeared to be valid. However, the accuracy improvement was neither complete nor all encompassing.

AFIT Designator

AFIT-GSM-LAS-96S-6

DTIC Accession Number

ADA319050

Comments

Presented to the Faculty of the Graduate School of Logistics and Acquisition Management of the Air Force Institute of Technology

Share

COinS