Nicolas Debaene

While working at BNP Paribas, I developed a statistical and machine learning model to detect potential churners within clients of the bank. I created a method based on time series forecasting to predict the amount of forex transfer each client would request in the next year. The bankers could then take steps to retain the client. I developed a statistical time series forecasting (ARIMA model) and used deep learning auto-regressive models to achieve better performance.

• Developed a product categorization model (NLP) from scratch using machine learning and Python to automatically classify products from the eCommerce application thanks to their description and title.
• Assigned the model to automatically classify four million products sold in the company's eCommerce application in eight different countries in Latin America in 400 different categories (from fresh food, like fruits or meats, to beauty products).
• Productionized the model that runs automatically when onboarding new products.
• Developed a matching tool using natural language processing to match new onboarding products to products already existing in the catalog of the eCommerce application (to use standard information from products already verified from the catalog).

• Designed and developed an accurate, useful, and stable credit risk model to score customers with existing loan histories on their ability to repay future loans
• Created a second model to score prospects with no historical loan data
• Developed an optimal credit granting strategy balancing default and acceptance rate depending on the company's tolerated default rate