ML Pipeline Exploiting Known Attribute Value Uncertainties
Overview
Real-world data, as measured or observed, almost always contains some amount of uncertainty, including uncertainty in case-by-case values of attributes. AI algorithms are currently not equipped to make use of the uncertainty of the attribute values if these are known (estimated) and given alongside the data itself. This pipeline and modified algorithms for supervised machine learning can make use of such data and improve their predictive performance. Supervised machine learning methods, primarily decision tree-based (random forest and boosting techniques), are extended and modified to take into account known case-by-case attribute value uncertainty. An example use case is astronomical data that contains observations and their uncertainties (case-by-case), e.g., for predicting whether an observed object is a quasar or not.
Outcome
An easy-to-use supervised machine learning pipeline based on modified ML algorithms that can improve predictive performance by exploiting known attribute value uncertainties.
Service Card
| Category | AI development and deployment |
| Domain | Machine Learning, Uncertainty-Aware Modelling |
| Vertical | Cross-domain |
| Target users | Anyone Building ML Models |
| Service type | Pipeline |
| Outputs | Trained model. |
| Requirements | Own dataset with known feature value uncertainties. |
| Limitations | |
| Availability | Q3/2027 |
| Licensing | |
| Visual content | |
| Visual caption |