An Empirical Analysis on Effectiveness of Source Code Metrics for Aging Related Bug Prediction

Abstract

Reliability of any software plays a very important role to assess the quality of the software systems. Reliability analysis and evaluation during early phases of software development life cycle (SDLC) significantly help developer and analysts in the proper allocation of limited resources during testing and maintenance process. The goal of this work is to develop one model using the measurement of the internal structure of the software system i.e., source code metrics for predicting the reason of failure in software during continuously running for a certain time i.e., software aging. Aging-Related Bugs are related with failure during the execution of the software, that leads to degradation in quality, system crashing, misuse of resources, etc.. In this paper, seven different sets of software metrics, seven model training algorithms, one data sampling technique have been empirically investigated and evaluated for predicting aging-related bugs classes. The trained aging-related bugs prediction models are validated using 5-fold cross-validation techniques. The final observation of this experiment work is assessed over seven open source application software systems. The high-value of performance parameters confirm the predicting capability of data sampling, sets of metrics, and training algorithms to predict aging-related bugs classes.