Optimized ensemble machine learning model for software bugs prediction

• Published in: Innovations in systems and software engineering 2023-03, Vol.19 (1), p.91-101
• Main Authors: Johnson, Femi, Oluwatobi, Olayiwola, Folorunso, Olusegun, Ojumu, Alomaja Victor, Quadri, Alatishe
• Format: Article
• Language: English
• Subjects: Accuracy; Aeronautics; Algorithms; Artificial Intelligence; Classifiers; Computer Applications; Computer Science; Data collection; Datasets; Defects; Machine learning; Optimization; Principal components analysis; Regression models; S.i. : Intelligence for Systems and Software Engineering; Software; Software Engineering; Training; Utilization
• ISSN: 1614-5046; 1614-5054
• DOI: 10.1007/s11334-022-00506-x

Software accuracy and efficiency checks are becoming of paramount interest to system users before utilization. As a result, twenty-first-century programmers are consciously developing less buggy, highly efficient, and robust software with a higher degree of accuracy. Occasionally, undetected bugs in large software due to the complexity of codes and other associated parametric attributes cause hardware to malfunction. In this paper, an ensemble model of Logistic Regression and Extra tree classifier algorithms is deployed on parametric software attributes for the accurate classification and prediction of software bugs. The implementation was performed on different platforms (WEKA, MATLAB and PyCharm) to determine the rate of memory utilization, optimize prediction time, maximize the model’s efficiency and compare accuracy rankings among similar machine models. A publicly available software defects dataset from the National Aeronautics and Space Administration (NASA) containing 16,962 instances and 38 attributes for software defects prediction was collected, pre-processed and used in the implementation of this study. The collected data were vectorized, subjected to principal component analysis (PCA) for dimension reduction based on ranking values and divided in the ratio 3:2 for training and testing of the ensemble model classifier, respectively, on new sets of buggy software datasets. The result from the ensembled model showed a significant increase from 96.7–97.8% in the prediction accuracy of the un-vectorized dataset to vectorized dataset. An appreciable decrease in the prediction time (19.7 s) of the vectorized dataset was also observed against the initial time (26.9 s) recorded for the un-vectorized dataset. In addition, memory utilization for vectorized datasets increased during the training phase due to the number of bits but got reduced at the final testing phase of the software bug prediction. However, the overall accuracy of 97.8% recorded by the optimized ensemble model for buggy software prediction proved the model’s capability to accurately classify and predict buggy software with efficient memory utilization at optimal time duration.