What-If Tool

Load loan application data and trained model into WIT.

Segment data by demographic groups.

Analyze fairness metrics (e.g., disparate impact) across groups.

Use counterfactual analysis to determine the impact of changing applicant features (e.g., income, credit score) on the prediction.

Identify and mitigate biased features by adjusting model parameters or retraining with debiased data.

Load image data and trained model into WIT.

Visualize model predictions for misclassified images.

Use partial dependence plots to identify image features (e.g., edges, textures) that influence the prediction.

Modify image features and observe the resulting changes in the prediction.

Identify and address issues with training data or model architecture.

Load customer data and trained model into WIT.

Analyze feature importance to identify key drivers of customer behavior.

Use counterfactual analysis to simulate the impact of different marketing interventions on customer response.

Segment customers based on model predictions and feature values.

Optimize marketing campaign targeting based on WIT insights.

Load historical data and current data into WIT.

Compare model performance on both datasets.

Analyze feature distributions to identify potential data drift.

Use partial dependence plots to understand how the relationship between features and predictions has changed.

Retrain the model with updated data or adjust model parameters to adapt to the new data distribution.

Load customer review data and trained sentiment analysis model into WIT.

Analyze model predictions for reviews with ambiguous sentiment.

Use counterfactual analysis to modify review text and observe the resulting changes in sentiment score.

Identify and address issues with model training data or sentiment lexicon.

Retrain the model with improved data and lexicon.