In the IEEE Standard Glossary of software engineering terminology, software reliability is defined as ‘‘The ability of the software to perform its required function under stated conditions for a stated period of time’’ (Standards Coordinating Committee of the IEEE Computer Society 1991). With the rapid growth and increasing complexity of the software, the software reliability is hard to achieve. Reliability is one of the important and crucial aspect and attribute of the software quality. According to ANSI, software reliability is ‘‘The probability of failure free operation of a computer program for a specified period of time in a specified environment’’ (Quyoum et al. 2010). Software Reliability Growth Models (SRGM) has been used for predicting and estimating number of errors in the software. The primary goal of software reliability modeling is to find out the probability of a system failing in given time interval or the expected time span between successive failures. ML techniques have proved to be successful in predicting better results than statistical methods and can be used for prediction of software failures more accurately and precisely Malhotra and Negi (2013). These techniques envisages past failure data as input and quite less assumption is required for modeling of the software’s with complex phenomena. ML is an approach which is focused on learning automatically and allows computers to evolve and predict the system behavior based on past and the present failure data. Thus it is quite natural for software practitioners and researchers to know that which particular method tends to work well for a given failure dataset and up to what extent quantitatively (Aggarwal et al. 2006; Goel and Singh 2009; Singh and Kumar 2010a, b, c). In this study, we present an empirical study of various ML techniques such as ANFIS, FFBPNN, GRNN, SVM, MLP, Bagging, CFBPNN, IBK, Lin Reg, M5P, RepTree, M5Rules for predicting software reliability based on five industrial datasets. Thereafter, we investigate about the accuracy and performances of ML based models in predicting the Software Reliability when applied to past failure week data Zhou et al. (2010). We also performed a comparative analysis between cumulative failure data and inter failure time’s data to investigate the type of failure data more appropriate for reliability prediction and inferred that cumulative data yields better results and is always preferred over inter failure time’s data Tian and Noore (2005). The results show that correlation coefficient for cumulative data always yields positive linear relationship which shows strong correlation between the actual and the predicted values for reliability prediction. The rest of the paper is organized as follows. Section 2 includes the objectives of the study. Section 3 summarizes the related research work conducted on software reliability prediction. Section 4 provides overview about the research background. Section 5 focuses on the various research methodologies used in predicting Software Reliability. Section 6 includes the results of study, analysis and discussion of the results. Section 7, highlights the threats to validity and finally Sect. 8 concludes the paper. Followed by references.