From: Iterative guided machine learning-assisted systematic literature reviews: a diabetes case study
Average post hoc model performance | ||||||
---|---|---|---|---|---|---|
Data set | Prediction threshold for 1st iteration | Prediction threshold for 2nd iteration | Prediction threshold for 3rd iteration | Total articles | % of total human-reviewed articles needed to return 95% relevant articles | % of total human-reviewed articles needed to return 98% relevant articles |
1st SR review | 50.0% | 20.0% | 20% | 14,655 | 19.3% | 24% |
2nd SR review | 50.1% | 30.3% | 44% | 15,234 | 18.9% | 25% |
3rd SR review | 75.0% | 20.0% | 20% | 7,670 | 10.0% | 34% |
4th SR review | 70.0% | 27.5% | 19.5% | 1,820 | 30.0% | 41.8% |
Weighted average | 57.6% | 26.0% | 29.5% | N/A | 20.9% | 29.8% |