Authors:
Salvatore Alessandro Sarcià
1
;
Giovanni Cantone
1
and
Victor R. Basili
2
Affiliations:
1
Università di Roma Tor Vergata, Italy
;
2
University of Maryland, United States
Keyword(s):
Risk Management Process, Artificial Neural Networks, Experimental Software Engineering, Prior Probability, Posterior Probability, Bayes’ Theorem, Computational Software Engineering.
Related
Ontology
Subjects/Areas/Topics:
User Modeling
;
Web Information Systems and Technologies
;
Web Interfaces and Applications
Abstract:
This paper enhances the currently available formal risk management models and related frameworks by providing an independent mechanism for checking out their results. It provides a way to compare the historical data on the risks identified by similar projects to the risk found by each framework Based on direct queries to stakeholders, existing approaches provide a mechanism for estimating the probability of achieving software project objectives before the project starts (Prior probability). However, they do not estimate the probability that objectives have actually been achieved, when risk events have occurred during project development. This involves calculating the posterior probability that a project missed its objectives, or, on the contrary, the probability that the project has succeeded. This paper provides existing frameworks with a way to calculate both prior and posterior probability. The overall risk evaluation, calculated by those two probabilities, could be compared to th
e evaluations that each framework has found within its own process. Therefore, the comparison is performed between what those frameworks assumed and what the historical data suggested both before and during the project. This is a control mechanism because, if those comparisons do not agree, further investigations could be carried out. A case study is presented that provides an efficient way to deal with those issues by using Artificial Neural Networks (ANN) as a statistical tool (e.g., regression and probability estimator). That is, we show that ANN can automatically derive from historical data both prior and posterior probability estimates. This paper shows the verification by simulation of the proposed approach.
(More)