ML accuracy
- Danielle Rubas (Deactivated)
If our machine learning pipeline can make a correct prediction, then we can save time. We offer many different lenses on the correctness of the predictions of our machine learning pipeline.
Evolution of accuracy over time (average and per field)
Accuracy by field for a given time range
Document level accuracy
Â
Â
Average accuracy | An average of the accuracy over all fields. 80% accuracy for example means that 80% of the predictions were correct. Note that we only take predictions into account for which we received feedback. |
Field accuracy | A detailed overview of the accuracy per field. |
Fully accurate documents | The percentage of documents that were fully accurate. 60% in this chart means that of all documents seen, 60% were fully correct, so no corrections were made by the reviewer in the entire document. This graph only takes documents into account for which feedback is available. |
Document accuracy | This chart groups documents with a similar accuracy score in terms of percentage of fields predicted correctly. |
Precision (per field) | Precision is a measure that indicates whether the predictions we make are correct |
Recall (per field) | Recall is a measure that indicates how many of the required predictions we actually made |
F1 (per field) | F1 is the harmonic mean of Precision and Recall |
Confusion Matrix | The confusion matrix sets out the predicted against the golden labels to easily see which labels still have confusion. |