Logger4u: Predicting debugging statements in the source code

Abstract

Software logging is an essential programming practice that saves important runtime information that can be used later by software developers for troubleshooting, debugging and monitoring the software. Even though software logging has numerous benefits this practice is underutilized because of lack of any formal guiding principles to developers for making strategic and efficient logging decisions. Logging should be optimized because too much logging can cause performance overheads; sparse logging can leave out vital information that might give clues to developers about the real issues. In absence of any formal guidelines developers rely solely on their domain knowledge and experience while making logging decisions. In order to lessen this effort of making decisions we have proposed a machine learning based framework, Logger4u for if-block logging prediction. We extract and use 28 distinctive static features from the source code helpful in making well informed logging decisions. We use Support Vector Machine (two variants, 1 linear and 1 RBF kernel based) models, Multilayer Perceptron with back propagation model and Random forest model in our work. Our approach gives encouraging results for if-block logging task. The accuracy achieved by the Linear SVM, MLP, Random Forest and kernel SVM are 73.05%, 74.62%, 79.84% and 81.22% respectively