Investigating Implications of Metric Based Predictive Data Mining Approaches towards Software Fault Predictions

Abstract

Context: Since 1990, various researches have been working in the area of software fault prediction but yet it is difficult to assess the impacts and progressive path of this research field. Objective: In this research work, author’s major objective is to investigate the context and dimensions of research studies performed by different researchers in the area of software fault prediction. This work also focuses on presenting a well defined systematic view of their findings and suggestions after a critical examination of all major approaches applied in this key research area. Method: This research work includes 112 total manuscripts published between 2009 and 2014. These studies are gathered from a pool of total 587 manuscripts. The selection criteria for these manuscripts are title, keywords and citation of that paper. Result: The results of this investigation shows that most of the research work related to software fault prediction have been performed on available data set from NASA repository. Most of the research work performed is basically confined to analysis or comparative study of various machine learning techniques based on their classification accuracy. Various research work published doesn’t exhibit clearer representation of any specific prediction model. Conclusion: Still after years of development, there is a huge gap between the industry requirement and the research being performed by different researchers in the field of Software fault prediction. A better collaboration between industry academia is still required. This research work represents a critical investigative approach towards finding the exact gaps to be filled and explored more authentic future research areas in this field. All result finding have been critically examined and compared with existing literature work for better understanding and deep insight over identifying the major strengths of chosen research field.