Mathematics, Engineering and Computer Science Undergraduate Theses

Browse

Recent Submissions

Now showing 1 - 5 of 150
  • Item
    Machine Learning for Sports Betting
    (2024) Rugg, Hank; Wendt, Theodore; Fasteen, Jodi; Smillie, Mark
    With sports betting becoming more widely legal, the use of machine learning algorithms for improving an individual’s odds of placing successful sports bets has increased. In general, applying machine learning algorithms comes with challenges such as data selection, feature engineering, and dealing with time series data. In the context of gambling, it also comes with ethical considerations such as the use of such models to gamble, the accuracy of the model, and the transparency of the model. This research focuses specifically on predicting the total combined score of NBA games. This is directly applicable to the Over/Under bet – “Over” if you believe the combined total score will be above the number set by the sportsbook and “Under” if you believe the combined score will be less than the number set by the sports book. The goal of this research is to create a machine learning model that can accurately predict the total combined score of NBA games.
  • Item
    Evaluating Predictors for the 2022 World Cup Using Decision Trees and Random Forests
    (2024) Morgen, Nicole; Fasteen, Jodi; Cline, Kelly; Morris, Jeffrey
    It is the goal of this project to accurately predict the results of the 2022 Qatar World Cup and develop a sophisticated model that can be applied to future World Cups. The FIFA 2022 World Cup was home to exciting upsets, devastating losses and unexpected results. FIFA rankings and past performance statistics were insufficient predictors for results, advancements and performances in the World Cup. In this project, machine learning algorithms and predictors beyond rankings will be used to predict the results of the Qatar 2022 World Cup. The FIFA ranking prediction method will serve as a baseline for accuracy. The two machine learning algorithms that will be considered for this project are decision trees and random forests, the latter of which can determine the validity of various parameters. The random forest algorithm created a more effective model for predicting the results of World Cup matches, exceeding the FIFA baseline accuracy by nearly 20% on individual games and by 6% in terms of team advancements. Some of the better predictors for World Cup matches were win proportions, FIFA points, the standard deviation in the age of players on the team and the average number of goals scored per game. While the random forest algorithm was a better predictor of games than both the FIFA baseline and decision tree models, it did not entirely accurately forecast the knockout bracket of the 2022 Qatar World Cup.
  • Item
    Beating the Pandemic: A Reinforcement Learning Approach to Cooperative Board Game Strategy
    (2023) Jarman, Anna; Wendt, Theodore; Fasteen, Jodi; Morris, Jeffrey
    Pandemic: Hot Zone - North America is an intricate board game in which players work together to treat disease and research cures for those diseases [3]. The game is part strategy and part luck. In this project we attempt to create an algorithm that will provide the best strategy for playing Pandemic. First, an optimization approach is taken in which we define na¨ıve strategies and analyze their effectiveness. Then, a reinforcement learning approach is taken in which we train an agent to play Pandemic in the same way humans learn to interact with our environment. The implementation of reinforcement learning in the context of this project is described in depth, including attempts to deal with a large state space and the use of hierarchical reinforcement learning. The results from this implementation are discussed and plans for future research and improvement are devised.
  • Item
    Understanding the Mathematics Behind Cryptography
    (2021) Kyriacos, Nicolas; Wendt, Theodore; Ayers, Kimberly; Scott, Shaun
    Today the internet is being used more than ever in order to store, transfer and access information. As a result, our private information is vulnerable to malicious parties who wish to use it for their own personal benefit. Cryptography uses mathematics in order to hide our private data so that it cannot be accessed by anybody other than ourselves. We begin this paper by exploring the most basic kinds of cryptosystems such as shift transformations, linear transformations and affine transformations as well as the how these cryptosystems are cracked using frequency analysis. We then take a look at how these concepts translate to enciphering matrices. Furthermore, we explain the differences between private and public key cryptography and investigate the RSA cryptosystem which is one of the most common types of public key cryptosystems in use today.
  • Item
    Comparing retirement investing with fixed and life-cycle funds using a variety of asset classes
    (2021) Drinville, Trevor; Cline, Kelly; Wendt, Theodore; Larsen, Peter
    There are many different studies that both reinforce and discourage the use of life-cycle funds for retirement investing, but a majority of these studies only look at life-cycle funds that include the three main asset classes, which are stocks, bonds, and cash. This study compares fixed funds, four different types of life-cycle funds, and the use of three other non-traditional asset classes in the life-cycle funds. The three non-traditional classes which we add to the study are real estate, commodities, and high yield bonds. We evaluate these different portfolios based on the average rate of return and their level of risk, by conducting random simulations based on historical data. Our simulations found that the use of non-traditional assets (commodities, real estate investment trusts, and high yield bonds) can be useful in a portfolio as they produce a higher mean rate of return than the portfolios with traditional assets (bonds and cash) that we studied. The downside of the use of non-traditional assets in a portfolio is that on average they are riskier than the traditional portfolios. In the study, we tested two portfolios that were focused primarily on high yield bonds, which showed that high yield bonds are a valuable asset class. These two portfolios are the only ones that have a mean rate of return that is indistinguishable from our 100% stock portfolio. They also have a much lower risk than the all-stock portfolio which is very rare to see.