International Journal of Computer Applications (0975 – 8887) Volume 116 – No. 8, April 2015 12 Analysis Receiver Operating Characteristics of Software Quality Requirement by Classification Algorithms Dhyan Chandra Yadav Research Scholar, Shri Venkateshwara University, Gajraula, Amroha (U.P.) Saurabh Pal Head, Dept. of MCA, VBS Purvanchal University. Jaunpur (U.P.) ABSTRACT Requirement engineering has an important role in software project development. Quality maintenance is the major factor in software industry. Requirement continues increases in the software market at different economic status with high class quality. The quality of software project development depend on technical performance but generally a technical problem run in project development known as duplicity. Duplicity is software bug which create problem in development. Data mining generate technical help in analysis of problematic area. In this paper we proposed the analysis of receiver operating characteristics of software defect related attribute data object and also analysis cost/benefit population, target confusion matrix and classification accuracy by zeroR, oneR and Prism algorithms of data mining. General Terms Data Mining, Classification algorithms, Software Engineering, Weka Tool. Keywords Data Mining; Classification: zeroR ,oneR and Prism; ROC; Weka. 1. INTRODUCTION Williams [1] discussed that duplicity is a technical problem in software project development life cycle. Duplicity is closed as more spaces between codes, change code in another location and error down line. All the technical problems mentation in a report known as problems report .If any bug reported in a problem report but it is already covered by another problem report this happening is known as duplicity of bug. Duplicate bug is created in any phase testing and it is some time automatically created in the coding implementation or phase testing by the help of data mining. It is easily classified and analyzed in the software engineering domain. Tiwari and Chaudhary [2] introduced about Data mining. Data mining provide facility in analysing data from different useful information. For example: Data Mining analysed bug or no bug in software project from related information. Data mining classifying the problems by zeroR, oneR and Prism algorithms and get accuracy of related attributes. Holte [3] discussed ZERO-R is an in consequential classifier, but it provides a minor certain presentation of a software defect database which should be suggestively better quality of software. It provides a sensible test on how glowing bug/no bug class can be expected without bearing in mind the other attributes. It is work as a lower bound for software defect data set and checks the instances values with class prediction. zeroR categorized numeric prediction problem in training data. In Weka [4] oneR is the second simplest classification model. It is planned for insignificant facts. It products modest instructions grounded on a particular quality. It generates one level decision tree for software defect and also count how each software defect class appears. OneR makes a rule for each Software defect attribute in training data and also calculates the error rate by this rule. Hong and Tseng [5] apply prism algorithm which has the impression of evidence improvement in its place of entropy as ID3. Quality respected couples in relationships of information theory, can be supposed of separate communications. The quantity of evidence improvement about a happening in a communication is defined as: Evidence improvement is selected for recounting a class with a larger importance. The mission of the prism algorithm is: a) Calculate the probability of incidence of the arrangement for each chooser. b) Choose the chooser for which probability incidence is a maximum then create a subset of the preparation set c) Repeat step 1 and 2 for this subset until it contains only instances of class classification. d) The multifaceted law is combination of all the choosers used in creating the comparable sub selection. e) At implementation set, remove all instance enclosed by multifaceted law. f) Reprat steps 1-5 until all occurrences of class classification have been detached. Swets and John [6] discussed that in figure a receiver operating characteristic (ROC) is a graphical plot that demonstrates the presentation of a binary classifier scheme as its judgment threshold is diverse. The arc is created by trickery the true positive rate beside the false positive rate at many threshold settings. The true positive rate is also known as sensitivity or recall in machine learning. The false positive rate is known as the fall out and can be calculated as specificity. The ROC curve is thus the sensitive as a function of fall out. In general, if the probability circulations for both discovery and false fear are known the ROC curve can be produced by trickery the cumulative distribution function of the detection probability in the y-axis versus the growing circulation function of the false arm possibility in x-axis.