Background: Software fault prediction is an important task to improve the quality of software. It reduces the time and complexity between modules. Future software faults depend on previous faulty data. Statistical Analysis: The prediction models are created using method level metrics available from NASA datasets of structure oriented CM1 and PC1 datasets and object oriented KC1 and KC2 datasets. These metrics were applied in different classifier like Naive Bayes, J48, K-Star and Random forest to identify the best classifier for small dataset and large dataset based on both structure and object oriented method level metrics. Findings: This paper gives the study and analysis of various methodologies used for predicting faults in both structure and object oriented software. Based on the study, Naïve Bayes is utmost suitable for small datasets and random forest is suitable for large datasets based on the evaluation done by us using various methodologies driven by WEKA tool while equating precision, recall and accuracy. Application: This process of software fault prediction can be applied to E-Commerce applications, where accuracy requires much importance.