Search

Software is becoming more complex, and lengthier in size, and needs to be updated on a time basis. But the constant change in codes has a high chance of arising faults that adversely affect the system performance. To avoid this interruption, and to improve the overall performance of software, a fault prediction system is required. The nature of the fault depends upon various parameters, but it is also important to identify the role of parameters in fault generation. In this paper, four feature selection techniques (Correlation Coefficient, Fisher Score, LASSO, Recursive Feature Selection) are applied on three machine learning methods individually (Random Forest, Logistic Regression, SVM) using Python 3.8 to compare the result for highest performance. The result clearly shows that Random Forest gives the highest accuracy of 95% when combined with the Fisher Score feature selection technique. The outcome shows that rather than using all features of any data set, a selected feature should be used for better fault-free software performance. [ABSTRACT FROM AUTHOR]