An approach of attribute selection for reducing false alarms

Defect Prediction is one of the method in SQA (Software Quality Assurance) that attracts developers because it can reduce the testing efforts as well as development time. One problem in defect prediction is `curse of dimensionality´, as hundreds of attributes are there in a dataset in software repository. In this paper we try to analyze whether there is any way to remove more attributes after attribute selection and the effect of this reduction of attributes on performance of defect prediction. We found that False positive rate (False Alarms) is reduced by using our method of attribute selection, which in turn can be used to reduce the resource allocation for detecting defective modules.