Use cases
H2O Model Analyzer currently supports tabular and text data for binary classification tasks. The project is under active development, and work is in progress to extend support for other data types and problems.
Loan application
Imagine a bank customer who submitted a loan request but got rejected by the bank based on the recommendations made by a predictive model. To further understand the predictive model's decision-making process and, in particular, verify that no irregularities were committed and the model was not impacted by a drastic change in the input data, you can use H2O Model Analyzer to arrive at such an understanding.
Bias analysis
A domain expert can create a scenario interactively by changing "age" or "gender" and evaluate a model's sensitivity to such critical changes. Did such a change flip the predictive outcome?
Natural language processing (NLP) sentiment analysis
A user can check the vulnerabilities of the trained model against the train/challenger or evaluation dataset before and after the model is deployed in production.
To assess the reliability of an image classification model, users can employ automated augmentation analysis. However, this feature is not available in the released version. If you find the provided example relevant, please contact cloud-feedback@h2o.ai.
- Submit and view feedback for this page
- Send feedback about H2O Model Analyzer to cloud-feedback@h2o.ai