The current consumer credit risk models are usually based on private information of the borrower and are not overly dynamic with respect to changing market conditions. We propose to use machine learning methods to analyze more subtle patterns in consumer expenditures, savings, and debt payments, than can the prominent models for consumer credit-default such as a logit, discriminant analysis, and credit scores. We plan to utilize radial basis functions, tree-based classifiers, and support-vector machines, which lend themselves well to large data sets and complex interaction networks between credit-default probability, consumer transactions, and general consumer characteristics.

I worked at a CERN computing center in Italy doing data analysis and helped develop a Monte Carlo algorithm to compute energy deposition in resists during Electron Beam Lithography with the Quantum Nanostructures and Nanofabrication Group at MIT. I spent this summer working at AgilOne, a predictive marketing company, as a data science intern and I spent last semester leading the software division of the MIT Robotics Team.