Early Prediction of Software Fault-Prone Module using Artificial Neural Network

Abstract

Prediction of software modules into fault-prone (FP) and not-fault-prone (NFP) categories using software metrics allows prioritization of testing resources to fault-prone modules for achieving higher reliability growth and cost effectiveness. This paper proposes an Artificial Neural Network (ANN) model with use of Sensitivity Analysis (SA-ANN) and Principal Component Analysis (PCA-ANN) for dimensionality reduction of the prediction problem. In SA-ANN model, a non-linear logarithmic scaling approach is used to scale metrics values, which improves quality of ANN training, followed by sensitivity analysis to rank and choose top Sensitivity Casual Index (SCI) value metrics. In PCA-ANN model, PCA is used for reducing dimensions of the problem and then the reduced dimension data is scaled using logarithmic function followed by training and prediction by ANN model. Simulations are carried out for four benchmark datasets to evaluate and compare the classification accuracy of proposed models with existing models. It has been found that non-linear scaling has good effect on predictive capability and PCA-ANN model provides higher accuracy than SA-ANN model and some other existing models for four datasets.

5 Figures and Tables

Cite this paper

@inproceedings{Bisi2014EarlyPO, title={Early Prediction of Software Fault-Prone Module using Artificial Neural Network}, author={Manjubala Bisi}, year={2014} }