An empirical study of software reliability prediction using machine learning techniques

The applications of machine learning techniques have shown remarkable improvements for the prediction of software reliability than traditional statistical techniques. In this paper, we apply some well-known machine learning methods such as artificial neural networks, support vector machines, cascade correlation neural network, decision trees and fuzzy inference system to predict the reliability of a software product. The proposed models have been evaluated using mean absolute error, root mean squared error, correlation coefficient and precision. The 16 software life cycle databases have been used for empirical studies. These databases are extracted from data and analysis center for software. A comparative analysis is performed in order to determine the importance of each method to assess the capability of software reliability prediction models. We also observe that these models may use in reliability predictions and results may be more close to the reality and precision is very effective with varied real-life failure datasets. Finally we conclude that proposed approach is more precise in its prediction capacity having better capability of generalization.