DetecÃÃo de mÃdulos de software propensos a falhas atravÃs de tÃcnicas de aprendizagem de mÃquina

AUTOR(ES)
DATA DE PUBLICAÇÃO

2008

RESUMO

The success of software depends directly on its quality. Traditionally, formal methods and manual code inspection are used to assure it. Generally, such methods have a high cost and demand a lot of time. Therefore, tests should be carefully planned to avoid waste of resources. Nowadays, companies are looking for faster and cheaper ways to detect software defects. Even though, with all the advances in the last years, the development of software is still an activity that depends intensively on human effort and knowledge. Many researchers and companies are interested on developing a mechanism that will be able to, automatically, detect software defects. In the last years, machine learning techniques are being used in many researches aiming this objective. The present dissertation investigates and presents a study on the viability of machine learning methods application on detecting fault-prone software modules. Artificial neural network and instance-based learning techniques will be used in this task, having as source of information, software metrics taken from the Metric Data Program (MDP) repository of NASA.We will also present a number of improvements, proposed by this research, for some of these classifiers. As detection of defective modules is a problem that is cost sensitive, this dissertation will also propose a mechanism that is capable of measuring analytically the cost of each decision taken by the classifiers

ASSUNTO(S)

avaliaÃÃo de custos de testes instance-based learning detecÃÃo de mÃdulos propensos à falha mÃtricas de software nasa software metrics machine learning redes neurais artificial neural networks mdp aprendizagem de mÃquina ciencia da computacao nasa detection of fault-prone software modules mdp instancebased learning tests costs evaluation

ACESSO AO ARTIGO

Documentos Relacionados