Download as PDF, PPTX
Uploaded byFeng Zhang
Cross-project Defect Prediction Using A Connectivity-based Unsupervised Classifier
The document discusses a connectivity-based unsupervised classifier for cross-project defect prediction in software. It highlights the limitations of traditional supervised classifiers and presents the performance of a novel spectral clustering approach against both unsupervised and supervised methods across multiple projects. Results indicate that the proposed unsupervised method can compete with supervised classifiers and sometimes outperforms them in defect prediction tasks.
More Related Content
Similar to Cross-project Defect Prediction Using A Connectivity-based Unsupervised Classifier
Cross-project Defect Prediction Using A Connectivity-based Unsupervised Classifier
- 1. Cross-project Defect Prediction Usinga Connectivity-based Unsupervised Classifier Feng Zhang Quan Zheng Ying Zou Ahmed E. Hassan
- 2.
- 3. Training Defect prediction Past datato build the model
- 4. Training Target Past datato build the model New Defect prediction
- 5. Training Target Past datato build the model New Within-project defect prediction
- 6. Target Past data tobuild the model Historical data may not be available
- 7. Target Historical data maynot be available
- 8. Other projects astraining data Target
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.
- 16.
- 17.
- 18.
- 19.
- 20.
- 21.
- 22.
- 23.
- 24.
- 25. Unsupervised classifier Software metrics Defect data Software metrics Target project Defect proneness Training project How About UsingUnsupervised Classifiers? Heterogeneity
- 26. Unsupervised classifier Software metrics Defect data Software metrics Target project Defect proneness Training project How About UsingUnsupervised Classifiers? HeterogeneityInitial attempts using K-means were not very successful.
- 27. How About UsingUnsupervised Classifiers?
- 28. How About UsingUnsupervised Classifiers? Short distance
- 29. How About UsingUnsupervised Classifiers? Short distance
- 30. How About UsingUnsupervised Classifiers? Long distance Long distance
- 31. How About UsingUnsupervised Classifiers? Long distance Long distance
- 32. How About UsingUnsupervised Classifiers? Connections Connections
- 33.
- 34.
- 35. Far away indistance but may be connected !c
- 36. Far away indistance but may be connected !
- 37. Far away indistance but may be connected ! Connection is more important than distance.
- 38. Far away indistance but may be connected ! Are defective software entities connected to each other?
- 39.
- 40. Stronger Stronger Weaker Within-community andcross-community connections
- 41. Stronger Stronger Weaker Defective entitiestend to connect to other defective entities. Within-community and cross-community connections
- 42.
- 43. Consider each entity(file/class) as a node
- 44. Step 1. Computesoftware metrics
- 45. Step 2. Builda graph based on the similarity
- 46. Step 3. Makea bipartition on the graph
- 47. Step 4. Labelthe defective cluster Defective Clean
- 48. 17 lines ofR code is provided in the paper
- 49. Looks simple? Doesit really work?
- 50. Research questions RQ1. Howdoes the spectral clustering based classifier perform in cross-project defect prediction? RQ2. Does the spectral clustering based classifier perform well in within-project defect prediction?
- 51.
- 52. Equinox JDT LuceneMylyn PDE AEEEM (5 projects) Subject projects (Total: 26)
- 53. Equinox JDT LuceneMylyn PDE AEEEM (5 projects) CM1 JM1 KC3 MC1 MC2 MW1 NASA (11 projects) PC1 PC2 PC3 PC4 PC5 Subject projects (Total: 26)
- 54. Subject projects (Total:26) Equinox JDT Lucene Mylyn PDE AEEEM (5 projects) CM1 JM1 KC3 MC1 MC2 MW1 NASA (11 projects) PC1 PC2 PC3 PC4 PC5 PROMISE (10 projects) Ant Camel Ivy Jedit Log4j Lucene POI Tomcat Xalan Xerces
- 55. Classifiers for comparison(Total: 9)
- 56. Unsupervised 1. K-means clustering(KM) 2. Partition around medoids (PAM) 3. Fuzzy C-means (FCM) 4. Neural-gas (NG) Classifiers for comparison (Total: 9)
- 57. Unsupervised 1. K-means clustering(KM) 2. Partition around medoids (PAM) 3. Fuzzy C-means (FCM) 4. Neural-gas (NG) Supervised 1. Random forest (RF) 2. Naïve Bayes (NB) 3. Logistic regression (LR) 4. Decision tree (DT) 5. Logistic model tree (LMT) Classifiers for comparison (Total: 9)
- 58. RQ1. How doesthe spectral clustering based classifier perform in cross-project defect prediction?
- 59. NASA AEEEM PROMISE RQ1. How doesthe spectral clustering based classifier perform in cross-project defect prediction?
- 60. … … … NASA AEEEM PROMISE RQ1. How doesthe spectral clustering based classifier perform in cross-project defect prediction?
- 61. … … … NASA AEEEM PROMISE Average AUC Average AUC Average AUC RQ1. How doesthe spectral clustering based classifier perform in cross-project defect prediction?
- 62. … … … Average AUC Average AUC Average AUC NASA AEEEM PROMISE Rank classifiers (Scott-Knott Test) RQ1.How does the spectral clustering based classifier perform in cross-project defect prediction?
- 63.
- 64. Red text: Unsupervised Blue text: Supervised Rank1 Rank 2 Rank 3 Rank 4 RQ1. Results (cross-project)
- 65. Red text: Unsupervised Blue text: Supervised Rank1 Rank 2 Rank 3 Rank 4 RQ1. Results (cross-project)
- 66. Red text: Unsupervised Blue text: Supervised Rank1 Rank 2 Rank 3 Rank 4 RQ1. Results (cross-project) Our approach can compete with supervised classifiers under study, and sometime is even better.
- 67. RQ2. Does thespectral clustering based classifier perform well in within-project defect prediction?
- 68. RQ2. Does thespectral clustering based classifier perform well in within-project defect prediction?
- 69. 50% 50% RQ2. Does thespectral clustering based classifier perform well in within-project defect prediction?
- 70. 50% 50% AUCTraining Testing RQ2. Doesthe spectral clustering based classifier perform well in within-project defect prediction?
- 71. 50% 50% AUCTraining Training Testing Testing AUC RQ2. Doesthe spectral clustering based classifier perform well in within-project defect prediction?
- 72. 50% 50% AUCTraining Training Testing Testing AUC 50% 50% AUCTraining Training Testing Testing AUC … (500random splits, thus 1,000 evaluations) RQ2. Does the spectral clustering based classifier perform well in within-project defect prediction?
- 73. 50% 50% AUC Rank classifiers (Scott-Knott Test) Training Training Testing TestingAUC 50% 50% AUCTraining Training Testing Testing AUC … (500 random splits, thus 1,000 evaluations) RQ2. Does the spectral clustering based classifier perform well in within-project defect prediction?
- 74.
- 75. RQ2. Results (within-project) 1 Randomforest Gold
- 76. RQ2. Results (within-project) 1 2 Randomforest Logistic regression Spectral clustering Logistic model tree Naïve Bayes Silver Gold
- 77. 1 2 3 Random forest Logisticregression Spectral clustering Logistic model tree Naïve Bayes Fuzzy C-means RQ2. Results (within-project) Silver BronzeGold
- 78. 1 2 3 Random forest Logisticregression Spectral clustering Logistic model tree Naïve Bayes Fuzzy C-means RQ2. Results (within-project) Silver BronzeGold Our approach can achieve similar performance as supervised classifiers, except random forest.
- 79.
- 88.