الفهرس 
abstractOnly 14 pages are availabe for public view
 |  | from | 137 |  |  |
| | | from | 137 | | |
Abstract
This thesis presents Spotted Hyena optimization algorithm as a classifier for solving instable accuracy rates due to the imbalanced nature and huge dimension of software defect datasets. The algorithm is implemented and applied on within- projects and cross-projects of software defect prediction with similar dataset.
The thesis applies the proposed algorithm on with-in software defect prediction by gathering historical data from a project of software (training phase) and predicts defects in the same project or newer versions (testing phase) i.e. training and testing phases are applied on the same project. Moreover, cross- projects software defect prediction for similar dataset is experimented. In cross-projects, historical data of projects isn’t presented or insufficient to train and build Software Defect Prediction model. Therefore, the model is trained and developed on one project and applied for cross projects or other projects. Hence, a novel meta-heuristic optimization algorithm for predicting software defects to improve quality of new software projects or projects with a shortage in historical data via predicting software defects is presented.
Support and confidence of classification rules are utilized as a multi-objective fitness function which helps the spotted hyena optimizer to serve as a classifier by finding the most fit classification rules among initial individuals. The experiments are applied on software datasets of NASA such as KC1, KC2, PC3 and JM1.
The results shows that the proposed Spotted Hyena algorithm improves the classification accuracy issue of software defect prediction that is considered one of the most significant problems due to the shortage and heterogeneous in historical data. The Spotted Hyena Classifier gives accuracy rates better than popular data mining techniques such as Naïve Bayes (NB), Bagging, Boosting, Support Vector Machine (SVM), and C4.5. Moreover, experimental results show and discuss other performance measures like precision, recall and confusion matrix.
Finally, there are numerous different points to apply the Spotted Hyena optimization algorithm as a classifier in the future especially in the area of software defect prediction such as training the classifier after applying feature reduction methodology. Also applying the proposed classifier for cross-project defect prediction for heterogeneous or dissimilar datasets is an interesting point.