Software defect prediction using graph sample and aggregate-attention network optimized with nomadic people optimizer for enhancing the software reliability

Abstract

The major objective of Software Defect Prediction (SDP) is to detect code location where errors are likely to occur to focus testing efforts on more suspect areas. Therefore, a high-quality software is developed that takes lesser time without effort. The dataset used for SDP usually contains more non-defective examples than defective examples. SDP is an important activity in software engineering that detect potential defects in software systems before they occur. For that, this paper proposes a Software Defect Prediction using Graph Sample and Aggregate-Attention Network optimized with Nomadic people Optimizer for enhancing the Software Reliability (graphSAGE-NPO-SDP). Here, the data are taken from Promise Repository dataset and given to the pre-processing. The pre-processing is done by normalization techniques of Min-Max Scaling. After preprocessing, the features are selected under Univariate Ensemble Feature Selection technique (UEFST). The classification process is performed by graphSAGE. The classification results are classified as defect class and non-defective class. The performance metrics, like Accuracy, Execution time, F-measure, Precision, Root Mean Square Error, Sensitivity, and Specificity is examined. The proposed graphSAGE-NPO-SDP method attains higher accuracy 32.45 %, 36.48 % and 28.34 % when compared to the existing models: Complexity-based over sampling technique in SDP (COT-ACI-SDP), Classification Method for SDP utilizing multiple filter feature selection approach (MLP-SDP), Boosted WOA-SDP and hybrid model depending on deep neural network based for SDP under Software Metrics (DNN-GA-SDP) respectively.