Inheritance metrics feats in unsupervised learning to classify unlabeled datasets and clusters in fault prediction

Area of Science:

• Software Engineering
• Computer Science

Background:

• Software fault prediction (SFP) faces challenges due to the lack of labeled training data.
• Object-oriented inheritance metrics offer insights into software complexity and structure.

Purpose of the Study:

• To evaluate the effectiveness of inheritance metrics in classifying unlabeled software datasets.
• To develop a new mechanism for labeling clusters as faulty or fault-free in SFP.

Main Methods:

• Collected ten public datasets incorporating inheritance and CK (Chidamber and Kemerer) metrics.
• Applied K-means clustering and the Euclidean distance formula.
• Developed an averaging mechanism for cluster labeling.
• Evaluated performance using metrics like True Positive Rate (TPR), Recall, Precision, F1-score, and ROC.

Main Results:

• Inheritance metrics significantly impact SFP, particularly in classifying unlabeled data.
• The proposed average mechanism effectively labels clusters for fault prediction.
• The study demonstrated improved classification accuracy for instances.

Conclusions:

• Inheritance metrics are valuable for improving SFP, especially for unlabeled datasets.
• The developed cluster labeling mechanism is suitable for practical SFP applications.
• Quality assurance practitioners can utilize inheritance metrics for dataset and cluster labeling.