![A GENERAL SOFTWARE DEFECT-PRONENESS
PREDICTION FRAMEWORK [1]
Defect prediction framework :
Data pre-processor: Log-filtering
Feature selector: Forward Selection , Backward Elimination
Learning algorithms : Na誰ve Bayes, J48, OneR
2January2016
9](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-9-320.jpg)
![A GENERAL SOFTWARE DEFECT-PRONENESS
PREDICTION FRAMEWORK [1]
Small changes to data representation can have a major impact
Feature selection one attribute at a time is not a practical solution for
large datasets
Different learning schemes should be chosen carefully for different
datasets
There is no clear indication about which combination should be used
for a particular dataset
2January2016
10](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-10-320.jpg)
![HOW MANY SOFTWARE METRICS SHOULD BE SELECTED FOR
DEFECT PREDICTION?[2]
Five filter-based feature ranking technique
Methodology
Min-max normalization
Pair of each independent attribute and class attribute
Ranking the attribute
Subset selection (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, and 20)
2January2016
11](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-11-320.jpg)
![HOW MANY SOFTWARE METRICS SHOULD BE SELECTED FOR
DEFECT PREDICTION?[2]
Three metrics on average can be enough for building an effective
prediction model
Eliminating 98.5% of the available metrics improves the result
It is not confirmed that it will work with all datasets
2January2016
12](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-12-320.jpg)
![CHOOSING SOFTWARE METRICS FOR DEFECT PREDICTION: AN
INVESTIGATION ON FEATURE SELECTION TECHNIQUES[3]
Hybrid attribute selection approach
Feature ranking
Feature subset selection
Removal of 85% metrics can enhance the performance of the
prediction model
2January2016
13](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-13-320.jpg)
![RESULT AND DISCUSSIONS
2January2016
45
Comparison of
AUC values of
different methods
Date set Wahono
[4]
Abaei
[5]
Ren [6]
Lowest Highest
CM1 0.702 0.723 0.550 0.724 0.7946
KC1 0.79 0.790 0.592 0.800 0.8006
KC2 - - 0.591 0.796 0.8449
KC3 0.677 - 0.569 0.713 0.8322
KC4 - - - - 0.8059
MC1 - - - - 0.8110
MC2 0.739 - - - 0.7340
MW1 0.724 - 0.534 0.725 0.7340
PC1 0.799 - 0.692 0.882 0.8369
PC2 0.805 - - - 0.8668
PC3 0.78 0.795 - - 0.8068
PC4 0.861 - - - 0.9049
PC5 - - - - 0.9624
JM1 - 0.717 - - 0.7167
AR1 - - - - 0.8167
AR3 - - 0.580 0.699 0.8590
AR4 - - 0.555 0.671 0.8681
AR5 - - 0.614 0.722 0.925
AR6 - - - - 0.7566](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-45-320.jpg)
![RESULT AND DISCUSSIONS
Dataset Song [1] Wang [7] Jobaer
[8]
CM1 0.695 0.663 0.5500 0.680
JM1 0.585 0.678 - 0.6152
KC1 0.707 0.718 - 0.7244
KC2 - 0.753 - 0.7835
KC3 0.708 0.693 0.6037 0.7529
KC4 0.691 - - 0.7036
MC1 0.793 - - 0.6904
MC2 0.614 0.620 - 0.6847
MW1 0.661 0.636 0.7202 0.6577
PC1 0.668 0.688 0.5719 0.7040
PC2 - - 0.7046 0.7468
PC3 0.711 0.749 0.7114 0.7232
PC4 0.821 0.854 0.7450 0.8272
PC5 0.904 - - 0.9046
AR1 0.411 - - 0.6651
AR3 0.661 - - 0.8238
AR4 0.683 - - 0.7051
AR6 0.492 - - 0.5471
2January2016
46
Comparison of
Balance values of
different methods](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-46-320.jpg)
![REFERENCES
2January2016
48
[1] Song, Qinbao, Zihan Jia, Martin Shepperd, Shi Ying, and Shi Ying Jin Liu. "A general
software defect-proneness prediction framework." Software Engineering, IEEE Transactions on
37, no. 3 (2011): 356-370
[2] Wang, Huanjing, Taghi M. Khoshgoftaar, and Naeem Seliya. "How many software metrics
should be selected for defect prediction?" In FLAIRS Conference. 2011
[3] Gao, Kehan, Taghi M. Khoshgoftaar, and Huanjing Wang. "An empirical investigation of
filter attribute selection techniques for software quality classification." In Information Reuse &
Integration, 2009. IRI'09. IEEE International Conference on, pp. 272-277. IEEE, 2009.
[4] Wahono, Romi Satria, and Nanna Suryana Herman. "Genetic Feature Selection for
Software Defect Prediction." Advanced Science Letters 20, no. 1 (2014): 239-244.
[5] Abaei, Golnoush, and Ali Selamat. "A survey on software fault detection based on different
prediction approaches." Vietnam Journal of Computer Science 1, no. 2 (2014): 79-95.
[6] Ren, Jinsheng, Ke Qin, Ying Ma, and Guangchun Luo. "On software defect prediction using
machine learning." Journal of Applied Mathematics 2014 (2014).](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-48-320.jpg)
![REFERENCES
[7] Wang, Shuo, and Xin Yao. "Using class imbalance learning for software defect prediction."
Reliability, IEEE Transactions on 62, no. 2 (2013): 434-443.
[8] Khan, Jobaer, Alim Ul Gias, Md Saeed Siddik, Md Hafizur Rahman, Shah Mostafa Khaled,
and Mohammad Shoyaib. "An attribute selection process for software defect prediction." In
Informatics, Electronics & Vision (ICIEV), 2014 International Conference on, pp. 1-4. IEEE,
2014
2January2016
49](https://image.slidesharecdn.com/bsse0426-160102173459/85/Thesis-Final-Presentation-49-320.jpg)
More Related Content
Similar to Thesis Final Presentation (20)
Recently uploaded (20)
![AnyDesk Pro 3.7.0 Crack License Key Free Download 2025 [Latest]](https://cdn.slidesharecdn.com/ss_thumbnails/tighteningeverybolt-fosdem2025-250224092147-cc175839-250305170448-f3a5e54e-250307111927-ff339e62-thumbnail.jpg?width=560&fit=bounds)
![SketchUp Pro Crack [2025]-Free Download?](https://cdn.slidesharecdn.com/ss_thumbnails/tighteningeverybolt-fosdem2025-250224092147-cc175839-250305114942-10710f44-thumbnail.jpg?width=560&fit=bounds)
Thesis Final Presentation
- 1. Institute of Information Technology University of Dhaka SELECTION AND REPRESENTATION OF ATTRIBUTES FOR SOFTWARE DEFECT PREDICTION Supervised by Dr. Mohammad Shoyaib Associate Professor Presented by Sadia Sharmin BSSE-0426
- 2. CONTENTS Background Motivation Problem Specification Objectives of Research Literature Review Methodology Result Analysis and Discussion Future Work 2January2016 2
- 3. BACKGROUND Software Defect Any flaw or imperfection in a software work product or software process Software Defect Prediction An approach to find out the defected part earlier before testing/releasing the product 2January2016 3
- 4. AN OVERVIEW OF SOFTWARE DEFECT PREDICTION PROCESS 2January2016 4 Data Set Pre- processing Attribute Selection Testing Data Prediction Result Training Data Prediction Model Training
- 5. MOTIVATION Identifying the software bugs in an early stage Allocating the test resources efficiently Minimizing the cost of software development Improving the quality and productivity of software 2January2016 5
- 6. WHY NEED PRE-PROCESSING Noisy Data Outliers Missing value or Conflicting value Inconsistency 2January2016 6
- 7. WHY NEED ATTRIBUTE SELECTION Attributes are not equally important No standard set of attributes 2January2016 7
- 8. OBJECTIVES OF RESEARCH To find out how the existing pre-processing can be used with the attribute selection methods more efficiently. To survey the existing methods and propose a proper attribute selection method. 2January2016 8
- 9. A GENERAL SOFTWARE DEFECT-PRONENESS PREDICTION FRAMEWORK [1] Defect prediction framework : Data pre-processor: Log-filtering Feature selector: Forward Selection , Backward Elimination Learning algorithms : Na誰ve Bayes, J48, OneR 2January2016 9
- 10. A GENERAL SOFTWARE DEFECT-PRONENESS PREDICTION FRAMEWORK [1] Small changes to data representation can have a major impact Feature selection one attribute at a time is not a practical solution for large datasets Different learning schemes should be chosen carefully for different datasets There is no clear indication about which combination should be used for a particular dataset 2January2016 10
- 11. HOW MANY SOFTWARE METRICS SHOULD BE SELECTED FOR DEFECT PREDICTION?[2] Five filter-based feature ranking technique Methodology Min-max normalization Pair of each independent attribute and class attribute Ranking the attribute Subset selection (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, and 20) 2January2016 11
- 12. HOW MANY SOFTWARE METRICS SHOULD BE SELECTED FOR DEFECT PREDICTION?[2] Three metrics on average can be enough for building an effective prediction model Eliminating 98.5% of the available metrics improves the result It is not confirmed that it will work with all datasets 2January2016 12
- 13. CHOOSING SOFTWARE METRICS FOR DEFECT PREDICTION: AN INVESTIGATION ON FEATURE SELECTION TECHNIQUES[3] Hybrid attribute selection approach Feature ranking Feature subset selection Removal of 85% metrics can enhance the performance of the prediction model 2January2016 13
- 14. METHODOLOGY SAL: Selection of Attribute with Log filtering 2January2016 14 Pre-process the data with logarithmic filter Rank the Attribute Select the best set of attributes Build the predictor
- 15. PRE-PROCESSING 2January2016 15 ln (n + ) where = 0.01
- 17. ATTRIBUTE RANKING 2January2016 17 A1 A2 A3 A4 A5 An A1 0.564 A2 0.764 A3 0.685 A4 0.798 A5 0.892 . . An 0.789 Individual Balance value
- 18. ATTRIBUTE RANKING 2January2016 18 A1 A2 A3 A4 A5 An Individual Balance value A1 A2 A3 A4 A5 An A1A2 A1A3 . . A3A1 A3A2 . . AmAn Pair wise combination A1 0.564 A2 0.764 A3 0.685 A4 0.798 A5 0.892 . . An 0.789
- 19. ATTRIBUTE RANKING 2January2016 19 A1 A2 A3 A4 A5 An A1 0.034 A2 0.034 A3 0.456 A4 0.348 A5 0.784 . . An 0.789 Individual Balance value A1 A2 A3 A4 A5 An A1A2 A1A3 . . A3A1 A3A2 . . AmAn Pair wise combination A1A2 0.896 A1A3 0.734 .. .. A3A1 0.587 A3A2 0.669 .. .. AmAn 0.897 Pair wise Balance value
- 20. ATTRIBUTE RANKING 2January2016 20 A1 A2 A3 A4 A5 An A1 0.034 A2 0.034 A3 0.456 A4 0.348 A5 0.784 . . An 0.789 Individual Balance value A1 A2 A3 A4 A5 An A1A2 A1A3 . . A3A1 A3A2 . . AmAn Pair wise combination Pair wise Balance value Average Balance value for each attribute A1A2 0.896 A1A3 0.734 .. .. A3A1 0.587 A3A2 0.669 .. .. AmAn 0.897 A1 0.765 A2 0.534 A3 0.679 A5 0.987 A4 0.869 ... ... An 0.897
- 21. ATTRIBUTE RANKING 2January2016 21 A1 A2 A3 A4 A5 An A1 0.034 A2 0.034 A3 0.456 A4 0.348 A5 0.784 . . An 0.789 Individual Balance value A1 A2 A3 A4 A5 An A1A2 A1A3 . . A3A1 A3A2 . . AmAn Pair wise combination Pair wise Balance value Average Balance value for each attribute Average Balance Value = (Individual value + Average value of n pair)/2 A1 0.765 A2 0.534 A3 0.679 A5 0.987 A4 0.869 ... ... An 0.897 A1A2 0.896 A1A3 0.734 .. .. A3A1 0.587 A3A2 0.669 .. .. AmAn 0.897
- 22. ATTRIBUTE RANKING 2January2016 22 A1 A2 A3 A4 A5 An A1 0.034 A2 0.034 A3 0.456 A4 0.348 A5 0.784 . . An 0.789 Individual Balance value A1 A2 A3 A4 A5 An A1A2 A1A3 . . A3A1 A3A2 . . AmAn Pair wise combination Pair wise Balance value A1 0.765 A2 0.534 A3 0.679 A5 0.887 A4 0.869 ... ... An 0.897 Average Balance value For each attribute A5 0.887 A4 0.869 A10 0.765 A8 0.750 A9 0.696 ... ... An 0.523 Sorted Balance value in decreasing order A1A2 0.896 A1A3 0.734 .. .. A3A1 0.587 A3A2 0.669 .. .. AmAn 0.897
- 23. SELECT BEST SET OF ATTRIBUTES 2January2016 23 A5 A4 A10 A8 A9 ... ... An Ranking of Attributes Best Set of Attributes
- 24. SELECT BEST SET OF ATTRIBUTES 2January2016 24 A5 A4 A10 A8 A9 ... ... An Ranking of Attributes Best Set of Attributes
- 25. SELECT BEST SET OF ATTRIBUTES 2January2016 25 A5 A4 A10 A8 A9 ... ... An Ranking of Attributes Best Set of Attributes
- 26. SELECT BEST SET OF ATTRIBUTES 2January2016 26 A4 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887
- 27. SELECT BEST SET OF ATTRIBUTES 2January2016 27 A4 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887
- 28. SELECT BEST SET OF ATTRIBUTES 2January2016 28 A4 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887
- 29. SELECT BEST SET OF ATTRIBUTES 2January2016 29 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887 A4 2nd ranked
- 30. SELECT BEST SET OF ATTRIBUTES 2January2016 30 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887 A4 2nd ranked A5A4
- 31. SELECT BEST SET OF ATTRIBUTES 2January2016 31 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887 (previous) A4 2nd ranked A5A4 0.891 (new) Combined Balance value
- 32. SELECT BEST SET OF ATTRIBUTES 2January2016 32 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887 (previous) A4 2nd ranked A5A4 0.891 (new) Combined Balance value new value > previous value
- 33. SELECT BEST SET OF ATTRIBUTES 2January2016 33 A10 A8 A9 ... ... An Ranking of Attributes A5 Best Set of Attributes A5 1st ranked 0.887 A4 2nd ranked
- 34. SELECT BEST SET OF ATTRIBUTES 2January2016 34 A10 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891
- 35. SELECT BEST SET OF ATTRIBUTES 2January2016 35 A10 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891
- 36. SELECT BEST SET OF ATTRIBUTES 2January2016 36 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891 A10 3rd ranked
- 37. SELECT BEST SET OF ATTRIBUTES 2January2016 37 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891 A10 3rd ranked A5A4A10
- 38. SELECT BEST SET OF ATTRIBUTES 2January2016 38 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891 A10 3rd ranked A5A4A10 0.856 (new) Combined Balance value
- 39. SELECT BEST SET OF ATTRIBUTES 2January2016 39 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891 (previous) A10 3rd ranked A5A4A10 0.856 (new) Combined Balance value new value < previous value
- 40. SELECT BEST SET OF ATTRIBUTES 2January2016 40 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes A5A4 0.891 A10 3rd ranked Discarded
- 41. SELECT BEST SET OF ATTRIBUTES 2January2016 41 A8 A9 ... ... An Ranking of Attributes A5,A4 Best Set of Attributes Continue this process.
- 42. SELECT BEST SET OF ATTRIBUTES 2January2016 42 A5,A4,A9,A12,A7 Best Set of Attributes
- 43. PERFORMANCE MEASUREMENT SCALES 2January2016 43 Confusion Matrix Predicted Actual TP FN FP TN False Positive rate TruePositiverate 0 1 1 Area Under the ROC curve (AUC)
- 44. RESULT AND DISCUSSIONS Data set : NASA MDP repository and PROMISE repository Classifier : Na誰ve Bayes Performance Metrics : Balance , AUC (Area Under the ROC Curve) Programming Language : Java Machine Learning Tool : WEKA 2January2016 44
- 45. RESULT AND DISCUSSIONS 2January2016 45 Comparison of AUC values of different methods Date set Wahono [4] Abaei [5] Ren [6] Lowest Highest CM1 0.702 0.723 0.550 0.724 0.7946 KC1 0.79 0.790 0.592 0.800 0.8006 KC2 - - 0.591 0.796 0.8449 KC3 0.677 - 0.569 0.713 0.8322 KC4 - - - - 0.8059 MC1 - - - - 0.8110 MC2 0.739 - - - 0.7340 MW1 0.724 - 0.534 0.725 0.7340 PC1 0.799 - 0.692 0.882 0.8369 PC2 0.805 - - - 0.8668 PC3 0.78 0.795 - - 0.8068 PC4 0.861 - - - 0.9049 PC5 - - - - 0.9624 JM1 - 0.717 - - 0.7167 AR1 - - - - 0.8167 AR3 - - 0.580 0.699 0.8590 AR4 - - 0.555 0.671 0.8681 AR5 - - 0.614 0.722 0.925 AR6 - - - - 0.7566
- 46. RESULT AND DISCUSSIONS Dataset Song [1] Wang [7] Jobaer [8] CM1 0.695 0.663 0.5500 0.680 JM1 0.585 0.678 - 0.6152 KC1 0.707 0.718 - 0.7244 KC2 - 0.753 - 0.7835 KC3 0.708 0.693 0.6037 0.7529 KC4 0.691 - - 0.7036 MC1 0.793 - - 0.6904 MC2 0.614 0.620 - 0.6847 MW1 0.661 0.636 0.7202 0.6577 PC1 0.668 0.688 0.5719 0.7040 PC2 - - 0.7046 0.7468 PC3 0.711 0.749 0.7114 0.7232 PC4 0.821 0.854 0.7450 0.8272 PC5 0.904 - - 0.9046 AR1 0.411 - - 0.6651 AR3 0.661 - - 0.8238 AR4 0.683 - - 0.7051 AR6 0.492 - - 0.5471 2January2016 46 Comparison of Balance values of different methods
- 47. FUTURE WORK Cross-project defect prediction Using other publicly available datasets 2January2016 47
- 48. REFERENCES 2January2016 48 [1] Song, Qinbao, Zihan Jia, Martin Shepperd, Shi Ying, and Shi Ying Jin Liu. "A general software defect-proneness prediction framework." Software Engineering, IEEE Transactions on 37, no. 3 (2011): 356-370 [2] Wang, Huanjing, Taghi M. Khoshgoftaar, and Naeem Seliya. "How many software metrics should be selected for defect prediction?" In FLAIRS Conference. 2011 [3] Gao, Kehan, Taghi M. Khoshgoftaar, and Huanjing Wang. "An empirical investigation of filter attribute selection techniques for software quality classification." In Information Reuse & Integration, 2009. IRI'09. IEEE International Conference on, pp. 272-277. IEEE, 2009. [4] Wahono, Romi Satria, and Nanna Suryana Herman. "Genetic Feature Selection for Software Defect Prediction." Advanced Science Letters 20, no. 1 (2014): 239-244. [5] Abaei, Golnoush, and Ali Selamat. "A survey on software fault detection based on different prediction approaches." Vietnam Journal of Computer Science 1, no. 2 (2014): 79-95. [6] Ren, Jinsheng, Ke Qin, Ying Ma, and Guangchun Luo. "On software defect prediction using machine learning." Journal of Applied Mathematics 2014 (2014).
- 49. REFERENCES [7] Wang, Shuo, and Xin Yao. "Using class imbalance learning for software defect prediction." Reliability, IEEE Transactions on 62, no. 2 (2013): 434-443. [8] Khan, Jobaer, Alim Ul Gias, Md Saeed Siddik, Md Hafizur Rahman, Shah Mostafa Khaled, and Mohammad Shoyaib. "An attribute selection process for software defect prediction." In Informatics, Electronics & Vision (ICIEV), 2014 International Conference on, pp. 1-4. IEEE, 2014 2January2016 49
- 50. 2January2016 50