Novel XGBoost Tuned Machine Learning Model for Software Bug Prediction
Aashish Gupta, Shilpa Sharma, Shubham Goyal, Mamoon Rashid
Abstract
As internet users grow, the quantity of data available on the web increases with it. Virtually everything that needs human effort or human presence can be replaced by the Software. While developing an application it follows the Software Development Lifecycle (SDLC). Within the early stages of development, it's a compulsory task to take care of system or bugs to avoid wasting time and effort during initial development phase to forestall any runtime crisis. In this paper, we used the machine learning models – Logistic regression, Decision Tree, Random Forest, Adaboost and XGBoost as state-of-art models for four datasets of NASA-KC2, PC3, JM1, CM1. Later on, new model was proposed based on tuning the existing XGBoost model by changing its parameter namely N_estimator, learning rate, max depth, and subsample. The results achieved were compared with state-of-art models and our model outperformed them for all datasets. The authors believe that this research will contribute in correctly detecting the bugs with machine learning approach.