### Table 1. Performance comparison of image quality assessment models on LIVE JPEG/JPEG2000 database [13]. SS-SSIM: single-scale SSIM; MS-SSIM: multi-scale SSIM; CC: non-linear regression correlation coefficient; ROCC: Spearman rank-order correlation coefficient; MAE: mean absolute error; RMS: root mean squared error; OR: outlier ratio.

2003

Cited by 13

### Table I: Number of outliers

2000

Cited by 1

### Table 2. Trajectory Outlier Limits

"... In PAGE 6: ... The limits are a function of phase of flight. Trajectory segments with a calculated error outside of any one of the limits shown in Table2 were excluded from all segment statistics. The majority of the exclusions were due to large cross track errors.... In PAGE 7: ...head times of fifteen minutes, calculated with the baseline trajectory prediction algorithm, are shown in Fig. 4. The total number of level flight segments with data for a look-ahead time of fifteen minutes was 813, including outliers. Approximately 17 percent of the segments were excluded from the statistics based on the limits defined in Table2 . The mean and standard deviation for the along track error of the remaining 672 segments were -1.... In PAGE 9: ... 10. Climb data outliers were excluded from the statistical segments based on the limits defined in Table2 . All but two outliers were attributed to excessive cross track error.... ..."

### Table.7 Mean Absolute Errors At a class level there appears to be vague relationship between number of instances and the rate (successfully classified) (Fig 1), though the standard deviation of a linear regression is relatively high (Table 8) suggesting a marginal relationship at best. This result indicates that whilst higher instance populations may improve the quality of the classification, there are other strongly influencing factors that are impacting the classification. At a purely observational perspective two significant outliers (UTILITIES amp; OPERATINGSYSTEM, 713 and 724 instances respectively are very broad descriptors of the possible issues that could be classified within them. Consequently, it may be argued that in some cases the vague nature of the class includes issues (and consequently a large set of possible words to describe) that are sweepers for problems difficult to describe but

2004

Cited by 1

### Table I: Overlap of SVM outlier lists with SCP outlier list

2002

Cited by 1

### Table 7: Comparisons in terms of outliers

"... In PAGE 3: ... We can observe a better performance of D_F on all parameters. Table7 shows the amount of clipping introduced only on voiced and mixed frames, of a length greater than X frames, where X=0, 10, 40. As the values show, the D_F introduce a lower amount of clipping than the FVAD solution.... ..."

### Table 5: Comparison with outlier removed.

"... In PAGE 12: ... This problem could be eliminated if filtration requirements were included as an input to the benchmarking calculations. If this building were eliminated from the data set, then the correlation coefficients are as shown in Table5 . Again, the Siegel-Tukey test indicates that the differences are not significant.... ..."