A linear wrapper method for detection of atypical points in classification
Bibliographic record
Abstract
The detection of atypical data in a dataset, using a linear wrapper approach is the focus of this research. Atypical points are considered to be the misclassified points that the proposed algorithm (Atypical Sequential Removing: ASR) finds not useful to the classification task. They may include outliers and/or overlapping samples. The majority of the available atypical detection techniques apply a filter approach in which there is no requirement for the filter to be consistent with the classifier in use. The fastest available wrapper techniques, on the other hand, have a quadratic running time which is prohibitive in practice for sample subset selection. The approach presented in this research is a linear wrapper technique that, instead of using any predetermined criteria, uses only the classifier itself and a performance measure to identify atypical points in the data. As a result, it is expected to be more consistent with the classifier in use. Using a cross validation scheme, ASR manages to give a reliable test performance while identifying and ranking the atypical points in the whole dataset. To ensure that ASR does not remove informative misclassified points, Ada-boost was compared with S-boost (trained with the data without atypicals). The results showed that when a significant portion of misclassified points were removed from the training set, S-boost had a very close performance to Ada-boost. In the comparison between ASR and the Mahalanobis filter method, the results shows that ASR was more accurate in identifying atypical points, it was more consistent with the classifier in use by keeping its performance as high as the classifier with no removal from the training set, and it was able to remove 30% more points than the Mahalanobis filter. However, the assertions in the literature (removing some points from the training can enhance the performance of classifiers) were not confirmed for overall performance under the experimented linear wrapper. Experiments on 20 benchmark datasets and 7 classifiers show promising results and confirm that this linear wrapper method has some advantages and can be used for atypical detection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.007 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".