Machine Learning Model for Automated Software Testing and Defect Analysis in Early Maintenance Applications

Abstract

For cognitive operations that statistically resemble human actions, ML solutions are finding their way into an expanding variety of software applications. A lot of human work goes into testing this kind of software by coming up with appropriate picture, text, and voice inputs and then judging if the program is processing them as humans do. When bad behavior is detected, it's not always easy to tell whether it originated in the cognitive ML API itself or in the code that uses it. When organizing software testing, ML classifiers for module fault prediction are helpful. The majority of these studies primarily assess the ML classifier's performance based on its accuracy. The ML model develops bias, however, when the classifier is trained and tested on imbalanced datasets. Hidden accuracy is caused by biased ML models. Within this framework, this research suggests a method to improve ML classifier performance in forecasting software module failures, even when dealing with imbalanced datasets. First, there was a marked improvement in the efficiency of software testing; second, there was a notable improvement in the number of modules that were accurately identified as defective; third, there was an increase in the number of modules that were tested; but, there were also some negative outcomes: a decrease in overall accuracy; an increase in the number of false positives; a decrease in software testing efficiency; an increase in the number of modules that were tested; and an increase in the number of modules that were tested. That is why it is important to think about the trade-off that the recommended strategy imposes when organizing software testing tasks. Lastly, this paper suggests a method to assist managers in handling these trade-offs while taking resource limits into account.