Prediction of Faults in Embedded Software Using Machine Learning Approaches

Software industry owns the most modern technological developments in this era. A software should be working perfectly without any error before handed over to the customer. The process of this quality assurance should be cost effective, less time-consuming, and reliable. But in practical scenario, it takes time to predict, detect and rectify the faults in a software. There are many models to predict the faults of software but no single model can claim to be perfect. This paper studied seven Machine Learning (ML) algorithms: Ada-Boost, CatBoost, LightGBM, Random Forest, XGBoost, Ensemble using Stacking and Voting to predict faults for an embedded software. Experiments were carried out with six datasets and eight performance metrics: accuracy, sensitivity, specificity, F-measure, balance, AUC, MAE, and precision to observe the performance of the algorithms. Random Forest showed the best results for accuracy, precision and specificity for 35%, 25% and 20% test data ratio with prediction of 0.995, 1 and 0.996 respectively. On the other hand, Ensemble (Stacking) showed the best result for specificity for 35% test ratio with prediction of 0.996. These results proved the suggested model better than the previous ones with acceptable outcome.