This program simulates golf shots based on the users previous shots. The program uses K-means, and linear algebra concepts to predict the shots. The program was as apart of golfsimulation.net, that reached 200 unique monthly visitors. This was my first leap into programming and data science, and helped me tremendously in my graduate program applications.
Anders Baumann, Jan. 2022
This thesis consisted of evaluating the effectiveness of different Nautral Language processing (NLP) sentiment models in predicting the inflation expecations of economic agents based on the Federal Open Market Committee (FOMC) monetary policy statements. We provided two new fine-tuned LLMs trained specifically for deriving inflation sentiment of FOMC minutes and Bank of England Monetary Policy Reports. We compared with the SOTA LLM at the time base GPT-4o, and our fine-tuned BERT model outperformed this model in both evaluations. This model is freely available on Huggingface and can be accessed by clicking this text.
Tuan Nguyen, Anders Baumann, June 2024
This project is based on Spotify's REST API, and contains two programs. The first one tries to predict whether a song will be featured on the Billboard Hot 100 based on its audio features. The F1 score is 0.33, and was concluded to be difficult for even the more complex classification models. The precision score is 0.22, so it would not be feasible to use this in a real business setting. The other program recommend songs based on content-based filtering, again solely based on audio features given by the Spotify API.
Anders Baumann, April 2023
In this project, I applied Transductive Conformal Prediction (TCP) to a binary classification problem, specifically focusing on predicting customer churn in the credit card industry. Conformal prediction is a model-agnostic framework that provides uncertainty estimates by producing valid prediction sets, enhancing traditional machine learning models. The project demonstrates how conformal prediction can be used to make more informed decisions in high-stakes environments. By quantifying uncertainty and providing control over the error rate, conformal prediction ensures that model predictions are not only accurate but also reliable.
Anders Baumann, Aug. 2024
This notebook was made as apart of a class I took my first semester for my masters program. The program was designed for university administrators to use as a part of their decision-making progress in potential investments for their university to boost their ranking.
Anders Baumann, Nov. 2022