IMPROVING DEFECT PREDICTION USING COMBINATION OF SOFTWARE METRICS
Software defect is one of the interesting challenges in the software development process. This is because a defect in the software can cause a failure in a software or its components so that it does not meet the design in the software requirements document. Software defect can be detected by testing...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/66655 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Software defect is one of the interesting challenges in the software development process. This is because a defect in the software can cause a failure in a software or its components so that it does not meet the design in the software requirements document. Software defect can be detected by testing all system components, but this process can be a allocated resource issue, especially limited time and manpower. Moreover, the cost of overcoming defects in software requires large costs. Software metrics are metrics obtained from software development data in a repository. From this repository data can be extracted into complexity metrics, change metrics and profile metrics. Complexity metrics are source code-based metrics, while change metrics store data on changes to a source code during development, and profile metrics describe profile data from contributing developers. This metric data can be used to predict the vulnerability of a class to defects. To predict this defect, a model is needed to predict the defect. Classification is a model that is often used to build predictive models. The main problem faced is the prediction results are less accurate. To solve this problem, this study proposes using a combination of software metrics. To improve the prediction results, the missing value is handled before the modeling is carried out. The experimental results show that the classification model with a combination of metrics produces better predictive results with the highest recall value in a combination of three metrics (complexity metrics, change metrics, profile metrics v3 (with the handling of missing values replaced by the median value)) in the Logistic Regression model and model with the best AUC value obtained in a combination of change metrics and profile metrics v3 with the Random Forest model. |
|---|