"Analyzing Inconsistencies Across Financial Services Machine Learning Algorithms and Implementations"
During Summer 2024, I have been working on my student-initiated research project, approved by the Colgate University Division for Natural Sciences and Mathematics. This research analyzes inconsistencies prevalent across various combinations of machine learning algorithms and their implementations on financial datasets.
Short introduction to the research
This research aims to assess the inconsistencies in outcomes produced by different machine learning algorithms and their implementations within the financial services industry. It focuses on assessing the reliability of these algorithms in tasks such as financial fraud detection, credit scoring, and trading decisions. Inconsistencies can arise due to the use of different combinations of algorithms and toolkits for financial tasks, as well as the lack of standard operating procedures across the industry. By exploring these inconsistencies at both the algorithm and implementation levels, the research seeks to enhance the interpretability and reliability of machine learning applications in financial contexts, ultimately aiming to ensure fairer and more accurate outcomes.
My role as an Undergraduate Researcher
Some of the technical tasks I have executed are listed below:​
​
-
Built random forest, gradient boosting, neural network, and linear regression models using Scikit-learn, XGBoost, LightGBM, TensorFlow, and PyTorch to enable classification and valuation tasks on 40+ financial datasets with an average accuracy of 90%
​
-
Compared output data from different model configurations by creating utilities to convert model outputs into a standardized CSV format for analysis, and applying statistical tests using Statsmodels and SciPy with data visualization using Matplotlib, which enabled identification of inconsistencies across 90+ pairs of model configurations
​
As this is a student-initiated project, I have also written the research proposal and the final paper for this project after consulting with my faculty supervisor.