International Journal Of Engineering Research And Development e-ISSN: 2278-067X, p-ISSN: 2278-800X, www.ijerd.com Volume 13, Issue 12 (December 2017), PP.01-11 1 Optimized Support Vector Machine for Software Defect Prediction 1 M. Thangavel, 2 Dr. R. Pugazendi 1 Department of Computer Science, Government Arts College(Autonomous),Salem-7. 2 Department of Computer Science, Government Arts College(Autonomous),Salem-7. Corresponding Author: M.Thangavel ABSTRACT: An error, bug, flaw, failure, mistake or fault in a computer program or system that generates inaccurate/unexpected outcome or prevents software from behaving as intended is a software defect. A project team wants to procreate a quality software product with zero defects. High-risk components in a software project must be caught early to enhance software quality. Software defects incur cost regarding quality and time. This article investigates Support Vector Machine’s (SVM) classification accuracy for Software Defect Prediction (SDP) and proposes a new optimized MRMR and SVM with firefly algorithm. Keywords:- Software Defect, Software Defect Prediction (SDP), optimized Support Vector Machine Radial Basis Function (SVM – RBF), Firefly. Date of Submission: 20 -11-2017 Date of acceptance: 28-12-2017 I. INTRODUCTION Defects can be defined in disparate ways but are generally aberration from specifications or ardent expectations which lead to procedure failures. Defect data analysis of classification and prediction types to extract models describing significant defect data classes or predict future defect trends. Classification predicts categorical/discrete, and unordered labels, while prediction models predict continuous valued functions. Such analysis provides better understanding of software defect data [1]. Software Defect Prediction (SDP) is essential in software engineering. Predicting defect is a proactive process, characterizing defect types in software’s content, design and codes to produce a high -quality product. Predicting defects in a system testing phase especially functional defects are important in test process improvement [2]. Software teams tries to produce a zero defect product. Defect prediction leads to a high- quality product and quality assurance. Software defects prediction reduces software testing efforts by guiding testers through software systems defect classification. If a defective product goes to customers it leads to issues. A defect reduces software reliability. Predicting defects needs practice and knowledge. Hence, defect prediction is important in software quality and software reliability [3]. Feature selection is a data preprocessing activity extensively studied in the machine learning and data mining community. The goal of feature selection is selecting a features subset that reduces classifiers prediction errors [4]. Feature selection techniques are divided into wrapper-based approaches and filter-based approaches. The former involves training a learner during feature selection, while the latter uses data intrinsic characteristics for feature selection based on a metric without depending on training a learner. The advantage of a filter-based approach over a wrapper-based approach is its faster computation. But, if computational complexity is not a factor, then a wrapper-based approach is the best overall feature selection scheme regarding accuracy [5]. Classification finds a set of models that describe/distinguish data classes/concepts. The derived model is represented in forms like classification rules and decision trees. When classes are defined the system infers rules governing classification. Hence, a system should find a description of each class. A description should refer to the training set’s prediction attributes so that positive examples satisfy the description. A rule is correct if its description covers all positive examples of a class [6]. Many classification methods were suggested to build SDP models. In [7], an association rule classification method which gets a comprehensible rule set from the data is proposed. They compared CBA2 [8] with other rule based classification methods to see whether association rules based classification algorithms suit software fault prediction. They investigated performance of an association rule based classification method for