Information Technology Reference
In-Depth Information
In order to answer research question 3, we look at Tables 8. We should mention
here, since Alan and Catal [23] did not evaluate their works based on testing, and
training datasets from different software projects, we conducted these experiments
according to their algorithms proposed and published in their paper. As it can be
seen in Tables 8, our proposed model performed best in terms of error rate in all
datasets except (AR4-AR5). It also outperformed other methods based on FPR rate
in almost all datasets. FNR rate has not improved based on our proposed method
compared to others. According to the results, we can conclude that our proposed
method can performed well compared to other methods, so we accept hypothesis 3
as well.
Table 8
Results based on different training and testing Sets
Dataset
Method
FPR
FNR
Error
ACN
0.03
0.41
0.06
JM1 for training & CM1 for testing
(JM1-CM1)
ACF
0.02
0.44
0.05
FMR
0.02
0.48
0.05
ACN
0.00
0.71
0.08
JM1 for training & KC1 for testing
(JM1-KC1)
ACF
0.01
0.50
0.07
FMR
0.00
0.63
0.07
ACN
0.00
0.54
0.12
JM1 for training & KC2 for testing
(JM1-KC2)
ACF
0.02
0.44
0.11
FMR
0.00
0.48
0.10
ACN
0.02
0.39
0.04
JM1 for training & PC1 for testing
(JM1-PC1)
ACF
0.02
0.33
0.04
FMR
0.02
0.33
0.04
ACN
0.14
0.00
0.12
AR4 for training & AR3 for testing
(AR4-AR3)
ACF
0.09
0.00
0.08
FMR
0.02
0.00
0.02
ACN
0.11
0.00
0.09
AR4 for training & AR5 for testing
(AR4-AR5)
ACF
0.07
0.00
0.06
FMR
0.07
0.14
0.09
7
Conclusion
In this paper, we have evaluated the effectiveness of fuzzy clustering based on
majority ranking in predicting faulty software module as compared to other famous
learning methods. Our proposed method clusters the input data based on available
industrial measurement thresholds and then any test data is labeled based on the
shortest distance from the modules in three most similar clusters. The overall error
Search WWH ::
Custom Search