Electronic Theses and Dissertations

Date of Award

1-1-2025

Document Type

Thesis

Degree Name

M.S. in Engineering Science

First Advisor

Kasem Khalil

Second Advisor

Sakib Hasan

Third Advisor

Elliott Hutchcraft

School

University of Mississippi

Relational Format

dissertation/thesis

Abstract

With the rapid development of Artificial Intelligence (AI) technology, its application in the financial field has moved from theoretical exploration to practical deployment. Especially in stock market forecasting, machine learning (ML) and deep learning (DL) have significantly improved the performance of traditional statistical methods by analyzing massive historical data. However, the stock market is affected by multiple dynamic factors, including macroeconomic indicators (e.g., GDP, interest rates), company fundamentals (e.g., earnings reports, management changes), technical indicators (e.g., moving averages, trading volume), and unforeseen black swan events (e.g., political unrest, natural disasters). These nonlinear interactions of these factors makes stock price prediction a challenging task.

To address this challenge, researchers have proposed a variety of forecasting models. For example, Long Short-Term Memory Networks (LSTMs) excel in stock price trend prediction with their ability to capture time series dependencies; Support Vector Machines (SVMs) handle high-dimensional data through kernel functions for small sample scenarios; And Random Forests (RFs) reduce the risk of overfitting through integrated learning. However, traditional models have limited performance when dealing with emerging stocks or data-scarce scenarios. It will lead to poorly trained models and weak generalization capabilities as the lack of historical data for newly listed companies. For this reason, it has become a key validation tool to simulate the trading strategies on limited financial historical datasets.

In addition to structured data, market sentiment has an increasingly significant impaction on stock prices. Investor sentiment can be conveyed through unstructured text such as news headlines, social media discussions, and earnings conference calls, which in turn trigger a chain reaction of buying and selling behavior. For example, a negative news report may lead to panic selling, while a positive product release may push up the stock price. To quantify the emotional impact, we used pre-trained models in finance (e.g., FinBERT) to categorize the text for sentiment (positive, neutral, negative) and input them as features into a prediction model. Experiments show that the LSTM model incorporating sentiment features improves the prediction accuracy by 13% compared without applying Fin-bert. Despite the enhanced predictive power of sentiment analysis, its application still faces two major challenges: data noise and timeliness.

In summary, AI-driven stock price prediction is gradually evolving from a single model to a multimodal framework. By integrating time series analysis, sentiment calculation and regression validation, our future research can further explore the direction of heterogeneous data fusion, and interpretability enhancement. Such frameworks can not only provide decision support for investors, but also provide technical tools for regulators to identify abnormal market fluctuations.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.