Citation Context

...o find, resolve and verify detected defects. These activitiessare required for quality management. The cost of defect correction and re-testingshas positive relation with the latency of the detection =-=[19]-=-. In other words, howsmuch late the defect is detected, that much more defect correction and re-testingscost is. Therefore, defect prevention and the analysis of remaining defects are twosimportant te...

Citation Context

...liable results and predictions can be used forssimilar projects or development teams with similar environment.s24s2.3.4 Prediction Models with Process Data by Machine Learning MethodssFenton and Neil =-=[54]-=- have evaluated defect oriented software metrics andsstatistical models. They have specified that reliability can not be computed bysusing defect density because the defects which cause not working of...

Citation Context

...ult. The collection of data from a tool’s database is categorized as a thirdsdegree data collection technique since collection by extracting data from databasesis independent of real development time =-=[12]-=-. Since this situation causes somesissues in mapping product data into process data which will be analyzed tosunderstand the software development process, the most of the organizations cansnot use the...

Citation Context

...l analysis is insufficient when we have a complex and largesdataset. They have specified that the application of both manual and machineslearning analysis gives the most accurate results.sWeiss et al =-=[61]-=- have used the defects life-time phases gone through issue trackingstool as the attributes for defect fix effort prediction. They compared two types ofsNearest Neighbor approaches called as with (α-kN...

Citation Context

...en the different nodes using the idiomssthrough analysis and coding of the interviewss5. Control the number of conditional probabilities that has to be elicited using thesdefinitional/synthesis idiom =-=[38]-=-s6. Evaluate the Bayesian belief network, possibly leading to a repetition of (asnumber of) the first 5 stepss7. Identify and define the conditional probability tables that define thesrelationships in...

Citation Context

...istake” cycle. In most cases defects cause fault andsfailures but this is not a must.sDefects are crucial for the quality of the product since it shows thesnonconformance to the customer requirements =-=[17, 18]-=-. Less defective software issmore reliable and reliability is an attribute of quality.s8sDefect detection, correction and verification have cost in the project, becausessome effort is spent to find, r...

Citation Context

...ve recommended genetic algorithm technique to predict faultsproneness of software modules. He has used requirements and code metrics calledsas product related metrics for his research.sÇatal and Diri =-=[49]-=- have reviewed software defect prediction studies in assystematical way. They have separated the studies to categories before review.sThe review states that the studies with using class-level, process...

Citation Context

...ng same metrics with previously mentioned twosstudies. The performance of the models has been evaluated by MAE, MMRE andscomparison between minimum MAE and median values of data groups.s25sSong et al =-=[57]-=- have suggested association rule mining for defect correction effortsprediction. Apriori accuracy values such as mean, median and standard deviationshave compared with the ones of PART, C4.5 and Naïve...

Citation Context

...sissIn multivariate linear regression, several independent variables are used to predictsone dependent variable. The relationship between dependent variable andsindependent variables are investigated =-=[29]-=-.sPrincipal Component Analysis (PCA)sPCA decomposes a data table with correlated measurements into a new set ofsuncorrelated variables [30]. The importance of each component is expressed bysthe varian...

Citation Context

... between the accuracy results of the model of code metrics, the modelsof history metrics, and the combination of them. They use machine learningsclassification and regression techniques.sFenton et al =-=[55]-=- have suggested Bayesian Belief Networks machine learningstechnique as prediction model. Process data is given for this model, again.sHe et al [56] have generated models with J48 (C4.5), Naïve Bayes a...

Citation Context

...efect classification and defect count understandingsis easy. By analyzing representations, defect progress in future can be predicted,sdecision making are performed, and defect prevention is achieved =-=[27]-=-.sUnivariate analysis is carried out with the description of a single variable and itssattributes of the applicable unit of analysis. If the variable defect data was thessubject of the analysis, the r...

Citation Context

...]. But analysis and interpretation of software development process data areshard since software engineering is an area which is affected from multiple factors.sFor example, in some prediction studies =-=[7, 10]-=-, authors suffer from the difficultys4sof collecting process-related data and taking into account all relevant evidences tosgenerate a prediction model.sIn order to understand the context of the produ...

Citation Context

...erformed by using size metrics such as SLOC and FPs(function points). His factors are pointed out to the dependence of analysis resultssto development environment and applied processes.sOhlsson et al =-=[46]-=- have built prediction models by using Principal ComponentsAnalysis (PCA) and Discriminant Analysis (DA) methods. They have usedsproduct design metrics for prediction. And they have divided software m...

Citation Context

...vention.sThe meanings of the terms mentioned in this study are below;sCase study: A research strategy, an empirical investigation technique thatsinvestigates a phenomenon within its real-life context =-=[20]-=-. This researchstechnique is commonly used in software related studies.sClass attribute:Dependent variable in statistics that is used for classification, youshave to select one of your attributes manu...

Citation Context

...RT, C4.5 and Naïve Bayes approaches. Defectstype metric has been used as input data. Also, false negative rate, false positivesrate performance values have been reviewed for evaluation.sZeng and Rine =-=[58]-=- have estimated defect fix effort by using dissimilarity matrixsand Self Organizing Maps (Kohonen Networks) which is a type of NeuralsNetworks method. With this data mining technique the data have bee...

Citation Context

... or not,susing statistical methods or machine learning ones, using assets to collect processsenactment data.s2.3.1 Prediction Models without Process Data by Statistical Analysis MethodssKoru and Tian =-=[43]-=- have validated the relationship between complexity and defectscount metrics by using statistical hypothesis tests. They have investigated in theirsstudy how high complexity affects defect count.sSalm...

Citation Context

...h Zeng and Rine. They havesconcluded their study that prediction model gives accurate results for the projectsswhich have same software development processes like product line projects.sMenzies et al =-=[60]-=- have presented a case study that compares defect analysissresults between machine learning and manual analysis used human expertise.sODC (Orthogonal Defect Classification) technique has been used. Th...

Citation Context

...lieve this should not be the only waysto use such models.s1.2 Difficulty of Collecting Defect Data With Process EnactmentsIn recent years software defect data analysis has been a common research area =-=[7,s8, 9]-=-. But analysis and interpretation of software development process data areshard since software engineering is an area which is affected from multiple factors.sFor example, in some prediction studies [...

Citation Context

...-level andscomponent-level measures are not sufficient. Besides, machine learning methodssare suggested because they give better results than statistical analysis and expertsview methods.sAhsan et al =-=[50]-=- have conducted a study to estimate bug fix effort. R (Pearsonscorrelation coefficient), MAE (Mean Absolute Error), RMSE (Root Mean SquaresError), MMRE (Mean Magnitude of Relative Error) and RRSE (Roo...

Citation Context

... scraps.sBecause development environment has high impact on these models,they aresspecific to the examined project.s2.3.2 Prediction Models without Process Data by Machine Learning MethodssBoetticher =-=[47]-=- has suggested nearest neighbor machine learning method to groupsdata. He has used product related metric data to predict the class in terms of itssdefectiveness status in the software.s22sSivrioğlu a...

Citation Context

...se risks, analysis of defectsdata is required. Besides defect data investigation provides quality improvementsand prevents injection of new defects by application of preventive actions to thesquality =-=[4]-=-. CMMI’s Causal Analysis and Resolution support process area atsmaturity level 5 suggests selecting defect data for cause analysis [1, 5]. Percentagesof defects removed, defect escape rates and number...

Citation Context

...singsPCA, redundant attributes are composed and attribute number decreases bysproviding new attributes, andat the end more meaningful and explanatorysattributes can be obtained. Otherwise overfitting =-=[69]-=- problem is common insmachine learning techniques.sIn numeric scale, attribute data should be discretized before analysis to obtainsmore meaningful analysis results. Some machine learning classificati...

Citation Context

...g to your dependent variable and the toolsgives you a model to be used for the prediction purpose with its performancesevaluation values. Class attribute is called as classifier in some studies (i.e. =-=[21]-=-).sDefect:Software bug that causes an incorrect or unexpected result, or causessproduct to behave in unintended ways.sDefect open duration: The period that elapses from the detection and recordingsof ...

Citation Context

...has not includedsprocess enactment data. The dataset is the data of a completed software project. Atsthe end of the study they have suggested to use contextual data for more accuratesresults.sManzoor =-=[45]-=- has tried code metric to estimate defect fix time. But the estimationsresults have not been found promising. Manzoor has explained the reasons of thissinaccurate estimation. He has given 14 factors w...

Citation Context

...velopment teams of same project.s23s2.3.3 Prediction Models with Process Data by Statistical Analysis MethodssIt is slightly possible to find studies by using process data in literature. Jalote et als=-=[51]-=- have explained a defect prediction approach by performing quantitativesquality management and statistical process charts.sWahyudin et al [9] have presented a defect prediction model by using statisti...

Citation Context

...hessame time, they give several factors which are considered affecting defect repairstime and causing a lognormal distribution in repair rates because of the factors’smultiplicativeness.sSchneidewind =-=[53]-=- has explained the delay between fault detection and faultscorrection times with exponential distribution. To obtain this statistical empiricalsresult, MSE (Mean Square Error) values of three operatio...

Citation Context

...r study before applying machine learning techniques. The performancesresults of clustered dataset and not clustered will be compared.s2.4 Methods to Collect Process Enactment DatasTarhan and Demirörs =-=[65, 66]-=- have emphasized the importance of processsdifferences in software projects. They have defined and applied some assets suchs27sas Metric Usability Questionaire (MUQ), Process Execution Record (PER), a...

Citation Context

...stering in Weka.sThe third question was “Which approaches or analysis methods can our methodssupport?”,and we explained the approaches applied in Section 3 in detail.sWhen we think of cost of quality =-=[70]-=-, performing defect prediction approachscosts 10 person-days for a project that shows similar features with the project ofsthe case study 1B that has 296 defects detected. After applying the generated...

Citation Context

...chieve multiple process areas.sIn this context, we first performed a case study for searching for analysisstechniques to understand product defectiveness and affecting factors in a smallsorganization =-=[2]-=-. We applied various statistical and machine learning analysissmethods to our product data. By doing this, we collected defect related andsproduct related metrics in different data sets. At the end, w...

Citation Context

...lieve this should not be the only waysto use such models.s1.2 Difficulty of Collecting Defect Data With Process EnactmentsIn recent years software defect data analysis has been a common research area =-=[7,s8, 9]-=-. But analysis and interpretation of software development process data areshard since software engineering is an area which is affected from multiple factors.sFor example, in some prediction studies [...

Citation Context

...ty management and statistical process charts.sWahyudin et al [9] have presented a defect prediction model by using statisticalshypothesis with a combination of product and process measures.sDhiauddin =-=[8]-=- has generated a prediction model for testing phase in his mastersthesis. With this model he discovers the strong factors that contribute to thesnumber of testing defects by using statistical methods ...

Citation Context

...alculated according to the createdsdate and closed date information of the defect obtained from the issue trackingstool. That is to say, we set open duration attribute as class attribute in Weka Tools=-=[13]-=- during machine learning classification operation.sTo answer the questions, we first decided which indicators and metrics would besuseful for this study. Therefore, the Goal-Question-Metric (GQM) [14]...

Citation Context

...ore, the detection of cause and its place aresvisualized for process stakeholders.s2.1 Defect Prediction BasicssA "mistake" or "fault" can be committed to the software at any stage duringsdevelopment =-=[15]-=-. When it cannot be detected, it causes unintended work of thessoftware product.sDefect is a stage of the “mistake” cycle. In most cases defects cause fault andsfailures but this is not a must.sDefect...

Citation Context

...actment data of defect managementsprocess. This data set was manually obtained by using Process Execution Records(PER) and Process Similarity Matrix (PSM) assets.sPER (Process Execution Record) forms =-=[16]-=- was filled by interviewing withsprocess experts. PSM (Process Similarity Matrix) was filled bysmanuallyreviewing issue tracking tool.sWEKA tool's [13] clustering facility (on cluster tab) was used to...

Citation Context

...A scientific data mining discipline that concerns with thesdesign and development of algorithms that allow computers to evolve behaviorssbased on empirical data, such as from sensor data or databases =-=[24]-=-. Machineslearning aims to recognize patterns and learn. Then, make intelligent decisionssbased on data after learning. For this purpose some part of whole data is separatedsas training data and remai...

Citation Context

...ibute onsanother attribute are searched with this analysis. According to the characteristicss13sof our data set, t-test, Z-test, Chi-square, ANOVA tests are some of the appliedsstatistical techniques =-=[26]-=-.sDuring analysis, below steps are performed [26];sStep 1: Null hypothesis and alternative hypotheses are stated.sStep 2: Significance level is set.sStep 3: The probability value are obtained by using...

Citation Context

...e purpose of bivariate analysis is explaining. It looks forsthe correlations, comparisons, relationships and causes between two variables.sDuring bivariate analysis, the steps given below are applied =-=[28]-=-;sStep 1: The nature of the relationship whether the values of the independentsvariables relate to the values of the dependent variable or not is defined.sStep 2: The type and direction, if applicable...

Citation Context

...o compute distance of each query instance to all trainingssamples.s2.2.2.2 C4.5 Decision TreesGiven a set S of cases, C4.5 first grows an initial tree using the divide-andconquer algorithm as follows =-=[34]-=-:s• If all the cases in S belong to the same class or S is small, the tree is a leafslabeled withthe most frequent class in S.s• Otherwise, choose a test based on a single attribute with two or more o...

Citation Context

...ve validated the relationship between complexity and defectscount metrics by using statistical hypothesis tests. They have investigated in theirsstudy how high complexity affects defect count.sSalman =-=[44]-=- has presented a measurement framework for component orientedssoftware systems as his PhD thesis. He has generated statistical regression modelssto predict size and effort metrics. The independent var...

Citation Context

...ase in his mastersthesis. With this model he discovers the strong factors that contribute to thesnumber of testing defects by using statistical methods such as regression analysis.sGokhale and Mullen =-=[52]-=- have hypothesized a Laplace Transform of thesLognormal distribution model with defect repair times data in day unit. At thessame time, they give several factors which are considered affecting defect ...

Citation Context

...stem mode, defect category and SLOC (source lines of code) changed.sDefect severity, detection activity, system mode and defect category attributes cansbe considered as contextual metrics.sThaw et al =-=[59]-=- have performed a similar study with Zeng and Rine. They havesconcluded their study that prediction model gives accurate results for the projectsswhich have same software development processes like pr...

Citation Context

...They compared two types ofsNearest Neighbor approaches called as with (α-kNN) and without thresholdss(kNN). They used text mining for grouping the data before kNN analysis.s26sHassouna and Tahvildari =-=[62]-=- have improved Weiss’ study by adding 1. datasenrichment to infuse additional issue information into the similarity-scoringsprocedure, 2. majority voting to exploit many of the similar historical issu...

Citation Context

...st the similaritysthreshold to ensure that they obtain only the most similar matches and 4. binarysclustering to form clusters when the similarity scores are very low phases.sHewett and Kijsanayothin =-=[63]-=- have penned down a comprehensive studysregarding defect repair time prediction. Firstly, they have applied five differentsempirical machine learning approaches to two individual data sets with andswi...

Citation Context

...osed model and compared the results. Defectsdetected testing phase, defect severity, defect state and defect state update datesshave been used as input attributes for prediction models.sMenzies et al =-=[64]-=- have pointed the importance of the models of similar regionssthan global ones in empirical studies. Two tools called WHERE to clustersalgorithm that divides the data and WHICH learner to find treatme...

Citation Context

...r study before applying machine learning techniques. The performancesresults of clustered dataset and not clustered will be compared.s2.4 Methods to Collect Process Enactment DatasTarhan and Demirörs =-=[65, 66]-=- have emphasized the importance of processsdifferences in software projects. They have defined and applied some assets suchs27sas Metric Usability Questionaire (MUQ), Process Execution Record (PER), a...

Citation Context

...ions against the test set. Since, the data in the testing set already containssknown values for the attribute that you want to predict, it is easy to determineswhether the model's guesses are correct =-=[72]-=-. The splitting 66% of the data set forstraining set and remaining for test is a commonly used technique.sCross ValidationsThe original data set is randomly partitioned into k sets. Of the k sets, a s...