MSc Thesis Defense Announcement of Ahmed Shafeek (Abouhassan):"TREE-BASED APPROACHES FOR PREDICTING FINANCIAL PERFORMANCE"

 

Date: Wednesday August 17,2022 

Time:  3:00pm – 4:30pm 

Location: Lambton Tower, Room 3105 

Reminder: Attendance must be recorded on both the Microsoft Form and the attendance sheet

 

In this research, we utilized a novel unused approach; we trained a model to classify future borrowers’ financial/repayment performance using nothing but their demographics and psychographics without including any prior financial history in the training data. The dataset we used is a large proprietary dataset of a US-based entity that required anonymity and de-identification of data and result specifics. We achieved an ROC-AUC of 86% in a binary classification target using Tree-Based Methods utilizing CatBoost API. We also experimented with a tri-class target and eventually applied our best model on one of the biggest publicly available financial datasets and received comparable results to the best research that was applied to this dataset with a fraction of the resources the original research utilized. The model we created for the data owner will be utilized for humanitarian purposes to plan and forecast future loan fund needs. Profitability, risk, and bottom lines are all non-factors for this specific group of people. Predictions will not be used to refuse loans to higher-risk populations, but rather to provide them with services and training that can better prepare them for success; a humanitarian approach to a financial problem. 

 

Keywords: Credit Scoring, Decision Trees, CatBoost, Financial Performance 

Internal Reader:    Dr. Ziad Kobti      

External Reader:   Dr. Hoda El Maraghy        

Advisor:                 Dr. Alioune Ngom 

Chair:                     Dr. Ikjot Saini