Cost-Sensitive Radial Basis Function Neural Network Classifier for Software Defect Prediction.

Area of Science:

• Software Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Software testing is crucial for delivering defect-free products, requiring efficient resource allocation and early defect detection.
• Identifying defect-prone software modules is key to optimizing the software development lifecycle and quality assurance efforts.
• Existing fault prediction methods aim to pinpoint areas within software more susceptible to defects.

Purpose of the Study:

• To propose an effective software fault prediction approach using a novel combination of machine learning models.
• To enhance the accuracy and efficiency of identifying defect-prone software modules.
• To improve the overall software testing process and resource allocation in software development.

Main Methods:

• A prediction approach combining a conventional radial basis function neural network (RBFNN) with an adaptive dimensional biogeography based optimization (ADBBO) model was developed.
• The proposed ADBBO-RBFNN model was evaluated using five publicly available datasets from the NASA data program repository.
• Performance was assessed using standard metrics to compare the model against existing predictors.

Main Results:

• The developed ADBBO-RBFNN model demonstrated significant effectiveness in predicting software modules prone to defects.
• Comparative analysis showed superior performance of the ADBBO-RBFNN approach over previously available predictors on the tested datasets.
• The model achieved notable results with respect to the considered evaluation metrics.

Conclusions:

• The proposed ADBBO-RBFNN classifier is an effective tool for predicting software defects, aiding in efficient resource management.
• This approach contributes to improving the software testing phase by enabling early identification of fault-prone modules.
• The study highlights the potential of integrating advanced optimization techniques with neural networks for robust software quality assurance.