ZZ Warner 208
303 College Street
Middlebury, VT 05753
View in Campus Map

Mariana Echeverria, 3:00 pm

¬†Breaking into the Black Box when Predicting Fraud Transactions

We can use machine learning models to predict fraud transactions, but how can we trust that the model is behaving as intended if we cannot trace back the prediction? Financial institutions wish to adopt machine learning to augment or even replace their traditional rules-based transaction monitoring systems. However, they fear that machine learning tools will create a black box for regulators and auditors. In this presentation, we will demonstrate how we have used machine learning algorithms to predict fraud transactions and performed model interpretability methods to test and validate that behavior of the models.

¬†Kailash Pandey, 3:45 pm

¬†Applications of machine learning models to find variable importance in predicting handwashing behavior in Bangladesh

Machine learning methods have been popular in the last decade after significant developments in computing power. However, they are very predictive but not easy to interpret like a simple regression model. It is still helpful to understand which are the most important variables when it comes to developing a model and extending it to policy applications. This independent study was about understanding different models and the tradeoff between computing power, interpretability, and predictability.

Sponsored by:

Contact Organizer

Nuceder, Jennifer