Reliable softwares are the need of modern digital era. Failure nonlinearity makes software reliability a complicated task. Over past decades, many researchers have contributed many parametric / non parametric software reliability growth models and discussed their assumptions, applicability and predictability. It concluded that traditional parametric software reliability models have many shortcomings related to their unrealistic assumptions, environment-dependent applicability, and questionable predictability. In contrast to parametric software reliability growth models, the non-parametric software reliability growth models which use machine learning techniques or time series modeling have been proposed by researchers. This paper evaluates and compares the accuracy of 2 parametric and 2 non parametric software reliability growth models on 3 real-life data sets for software failures.