Table 3
Results for kNN classification.
With equal priors | ||
Absolute | ||
bhb | other | |
BHB | 1008 | 235 |
OTHER | 347 | 871 |
Percentage | ||
BHB | 81.09 | 18.91 |
OTHER | 28.49 | 71.51 |
Completeness | Contamination | |
BHB | 0.811 | 0.422 |
|
||
With simple prior | ||
Absolute | ||
bhb | other | |
BHB | 743 | 500 |
OTHER | 245 | 973 |
Percentage | ||
BHB | 59.77 | 40.23 |
OTHER | 20.11 | 79.89 |
Completeness | Contamination | |
BHB | 0.598 | 0.412 |
|
||
With 2d prior | ||
Absolute | ||
bhb | other | |
BHB | 920 | 323 |
OTHER | 264 | 954 |
Percentage | ||
BHB | 74.01 | 25.99 |
OTHER | 21.67 | 78.33 |
Completeness | Contamination | |
BHB | 0.740 | 0.378 |
Notes. Top section: confusion matrix showing the results of kNN classification (k = 15) with threshold P(BHB) > 0.5. This is the combined result of ten independent classifications with resampling of the training and testing sets. The rows show the true class (according to Xue et al.), the columns show the classifier output class. When shown as percentages, the quantities in the rows should add to 100%, but the quantities in the columns in general do not. The completeness and contamination are also shown – and the contamination is corrected for the class imbalance using the simple prior. Middle section: the same confusion matrix, with output probabilities corrected for the simple prior. Bottom section: the results with output probabilities corrected for the prior probability as function of g magnitude and galactic latitude.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.