Matt Bezdek

Better Decision-Making with Explainable AI Models

So-called ‘black box’ models are popular, as in some cases they can provide boosts to accuracy over simpler and more transparent models. However, there is a cost to using a model that doesn’t reveal the relationship between its predictions and its input features. Luckily, even when using a black box model, there are techniques that can generate useful explanations of model predictions. Better yet, these techniques are available in free Python packages and work on a wide variety of model types. I will discuss applications of how to use explanation techniques to understand the predictions of otherwise opaque predictive models. I will show how meaningful explanations build trust in model predictions and improve the decision-making process through the integration of stakeholders’ subject matter expertise with the specific predictions of the model.