| Autor: Łukasz Radliński | 119 |
Strony: 119-137
pełen tekst
TECHNIQUES FOR PREDICTING DEVELOPMENT EFFORT AND SOFTWARE QUALITY IN IT PROJECTS
Summary
The most important dimensions in software project estimation are: development effort and software quality. Several predictive models have been proposed for these dimensions. Although some of these models provide useful input for decision makers, most of them are inherently limited for industrial use.
The aims of this study are to: (1) compare existing applications of methods and (2) select a best technique for building intelligent and practical models for development effort and software quality prediction. In recent years various techniques were used, which are based on statistics, machine learning, artificial intelligence and similar. Authors who used a single technique often report a success their studies, only sometimes additionally noticing threats in repeatability of their predictions in other environments. Other authors compare the accuracy of predictions obtained using different techniques.
The main problem in these analyses is the lack of straightforward confirmation of the usefulness of specific techniques in building predictive models. When one author finds that one technique performs the best in their study, another author obtains the best predictions using a different technique. Bayesian nets (BNs) appear to be the best suited approach to build predictive intelligent and practical model. This paper summarizes some applications of BNs in modelling different aspects of software engineering and discusses proposed Productivity Model for analysing trade-offs between effort, size and software quality.
pełen tekstTECHNIQUES FOR PREDICTING DEVELOPMENT EFFORT AND SOFTWARE QUALITY IN IT PROJECTS
Summary
The most important dimensions in software project estimation are: development effort and software quality. Several predictive models have been proposed for these dimensions. Although some of these models provide useful input for decision makers, most of them are inherently limited for industrial use.
The aims of this study are to: (1) compare existing applications of methods and (2) select a best technique for building intelligent and practical models for development effort and software quality prediction. In recent years various techniques were used, which are based on statistics, machine learning, artificial intelligence and similar. Authors who used a single technique often report a success their studies, only sometimes additionally noticing threats in repeatability of their predictions in other environments. Other authors compare the accuracy of predictions obtained using different techniques.
The main problem in these analyses is the lack of straightforward confirmation of the usefulness of specific techniques in building predictive models. When one author finds that one technique performs the best in their study, another author obtains the best predictions using a different technique. Bayesian nets (BNs) appear to be the best suited approach to build predictive intelligent and practical model. This paper summarizes some applications of BNs in modelling different aspects of software engineering and discusses proposed Productivity Model for analysing trade-offs between effort, size and software quality.