Research On Sotfware Defect Prediction Method Based On Probabilistic Relational Models

As progress of science and technology,computer technology has gradually become oneof the core technology of modern civilization. Its has widely used in all aspects of productionand living of the human, And plays an increasingly important role. At the same time,humandependence on the computer technology more and more in all respects, entertainment, office,travel, Medical, financial, aerospace, national defense. Computer technology has been animportant safeguard of high-speed developing of human civilization.Because of this increasingly dependent, as the basis for the computer to achieve a varietyof functions, The computer software quality directly affects the activities of the people inrelated fields. People began to research on how to improve the quality of computer softwarein the beginning of the computer technology. Generally speaking,we have two ways toimprove software quality: Reducing the defects in the software through a standardizeddevelopment methodology；Discovering and eliminating software defects by software testingin the development process.An important problem in the testing process is how to arrange test resources to reducethe cost of test and improve test results. If the test is not sufficient,there may be a lot ofdefects which would reduce the quality of the software in it; If the test is excessive, the costsof test will increase.So,we need to evaluate the quality of software testing.And the basis of theevaluate is the prediction of software defects characteristics,such as number, type, distribution.It can clearly be seen, the prediction of software defects plays an important role for the qualityof the software and the cost of development.The main work of this paper is proposing a software defect prediction method based onthe PRM through the study of software defect prediction techniques and ProbabilisticRelational Models.Probabilistic Relational Models is a Statistical relational learning methods based onbayesian networks. The standard Bayesian network is a graphical inference network based onprobability theory.Its network structure is a directed acyclic graph. The network noderepresents a property, Its range is all of possible values of the property;The directed edgeswhich connecte nodes represent the dependencies between the property. In this way,Bayesiannetwork is able to represent and reason the uncertainty knowledge.As an extension of the Bayesian network, Probabilistic Relational Models has same structure. The difference is that every node represents a class, And each class has its ownproperties,The models achieves the relationship between two classes or their properties by aseries slots.In this way, Probabilistic Relational Models improves the representation andreasoning capabilities for the uncertainty knowledge.It could handle multi-attribute problems.At the same time，because its theoretical basis is the probability theory, ProbabilisticRelational Models can also to predict uncertainty problem.In the software development process, The generation of software defects is anuncertainty problems.It is affected by many factors,such as development environment, humanfactors, software attributes and so on.These factors affecte the generation and distribution ofthe software defects in varying degrees.We need to consider all of these factors to predictsoftware defects accurately.So we need a tool which can describes predictes uncertaintyproblem.In this paper,We proposes a software defects prediction algorithm based on ProbabilisticRelational Models,It can inductives the dependencies between software defects and variousfactors by statistical methods, Predictes the mathematical characteristics of the defects in thesoftware.and guides testing in the software development process.To ensure that people canfind the software defects as much as possible and reduce development costs.And then,Achieve this algorithm, test and verify its effectiveness by experiment.