Leveraging Machine Learning to Combat Missingness and Error in Data
Bibliographic record
Abstract
ObjectivePoor quality data confounds efforts to link clients across datasets. To combat this, we have trialed an approach which seeks to identify linkage candidates using associations with related services. We utilised a machine learning (ML) approach to query linkage candidates’ dataset associations and make predictions about whether candidates are likely to be a client of the service being integrated for linkage. ApproachUtilizing a K-nearest neighbors algorithm, we trained a model using 15 variables to predict whether a linkage candidate is a client of a family service dataset. We then evaluated the model's success, and modified its native performance to minimise false positive matches. Subsequently, we tested the validity of these predictions as linkage criteria by employing blocking strategies and linking the service dataset to our linkage spine. The trained model was then used to identify correct links against the spine. ResultsThe evaluation of the machine learning model yielded promising results, with high accuracy (88.5%) and precision (95.5%). Testing the predictions as linkage criteria resulted in highly accurate links ranging from 96.0% to 98.7% across different blocking strategies. Despite some records failing to establish any links to the spine, rates of false positive matches remained low (0.7% to 3.1%). ConclusionMissingness and inaccuracy in data remains a key problem for data linkage, and a robust approach is required to resolve complex linkage cases. However, these findings suggest that machine learning can present novel options for a toolbox of many approaches to link problem records to a linkage spine.
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 | 0.072 | 0.167 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.004 | 0.004 |
| Science and technology studies | 0.002 | 0.002 |
| Scholarly communication | 0.005 | 0.007 |
| Open science | 0.004 | 0.004 |
| Research integrity | 0.003 | 0.004 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".