Cognitive Complexity and Graph Convolutional Approach Over Control Flow Graph for Software Defect Prediction

The software engineering community is working to develop reliable metrics improve quality. It estimated that understanding the source code accounts for 60% of maintenance effort. Cognitive informatics important in quantifying degree difficulty or efforts made by developers understand code. Several empirical studies were conducted 2003 assign cognitive weights each possible basic control structure software, and these are used several researchers evaluate complexity systems. In this paper, an effort has been categorize Control Flow Graphs (CFGs) nodes according their node features. our case, we extracted seven unique features from program, feature was assigned integer value evaluated through Complexity Measures (CCMs). We then incorporated CCMs’ results as a CFGs generated same based on connectivity graph. order obtain representation graph, vector matrix created graph passed Graph Convolutional Network (GCN). prepared data sets using GCN output built Deep Neural Defect Prediction (DNN-DP) (CNN-DP) models predict defects. Python programming language used, along with Keras TensorFlow. Three hundred twenty programs written talented UG PG students, all experiments carried out during laboratory classes. Together three skilled lab programmers, they compiled ran individual program detected defect/no-defect before categorizing them into different classes, namely Simple, Medium, Complex programs. Accuracy, Receiver Operating Characteristics (ROC), Area Under Curve (AUC), F-measure, Precision hyper-parameter tuning procedures approaches. experimental show proposed outperformed state-of-the-art methods such Nave Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF) evaluation criteria.