UPM Institutional Repository

New functional size convertibility models in FPA and COSMIC measurement methods


Ahmad Abualkishik, Abedallah Zaid (2012) New functional size convertibility models in FPA and COSMIC measurement methods. PhD thesis, Universiti Putra Malaysia.


Software functional size measurement is highly demanded and has gained wide adoption and acceptance in software organizations due to its benefits and wide applications in software project management. Function point analysis is the first method proposed by Albrecht, and has been maintained by the international function point user group. Function point analysis method is the most used measurement method globally. COSMIC method has been known as a second generation functional size measurement due to it is novel design. The method was designed to size a wider scope of functional domains, in particular, to measure real time systems and to alleviate the existing limitations of previous proposed methods. The need for conversion is driven by a method’s unsuitability for the task at hand, or its limitations, or it might be necessary because of the need to use the benchmark set of a particular domain. This is mainly, because function point analysis cannot size as many software functional domains as COSMIC, and because of some limitations surrounding function point analysis. The main problem with this change is to maintain the software organization’s ability to accurately convert their historical data measured by function point analysis to the corresponding value in COSMIC method. This thesis proposes a new theoretical model that converts the functional size measured by function point analysis to its corresponding COSMIC measures, at the level of base functional components of both methods, using the principles of probability based on in depth analysis of the type of transaction functions and its primary intent, processing logic forms and COSMIC method rules. The model was found to adequately convert all the tested applications precisely, in which it converts 97.7% of the whole dataset elementary processes into the estimated interval accurately. Most convertibility studies between the two methods undertook to convert the unadjusted function point to COSMIC size statistically. Two studies used the transaction functions size to obtain the corresponding COSMIC size, and found it more accurate than the type that uses the unadjusted size to obtain COSMIC measures. Accordingly, this thesis examines the accuracy of the two common statistical conversion types as well as the effect of function point analysis weighting tables and structural problems on its accuracy. Moreover, it proposes a new statistical conversion type that uses the number of files referenced by the whole elementary processes in a single application as a unit for prediction to estimate the corresponding COSMIC measures. Basically, two regression models have been used to compare the accuracy of the two statistical conversion types with the proposed type, based on the accuracy of fitting measures that uses the leave one out cross validation technique applied on one dataset. Also, four datasets from previous studies were used to further emphasize the obtained results. The two conversion types most often used were found to generate non-linear, inaccurate and violate the principle of measurement theory as scales transformation. The proposed statistical conversion type avoids the problems inherent in the other two types but not the non-linearity problem, and produced valid and highly accurate results over the tested datasets.

Download File

FSKTM 2012 24R.pdf

Download (1MB) | Preview

Additional Metadata

Item Type: Thesis (PhD)
Subject: Function point analysis
Subject: Software measurement
Call Number: FSKTM 2012 24
Chairman Supervisor: Assoc Prof Mohd. Hasan Selamat
Divisions: Faculty of Computer Science and Information Technology
Depositing User: Haridan Mohd Jais
Date Deposited: 04 Mar 2015 05:14
Last Modified: 04 Mar 2015 05:14
URI: http://psasir.upm.edu.my/id/eprint/33142
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item