[PDF]

Investigating whether news sentiment analysis improves machine learning model accuracy for the next day stock price direction prediction


Ziad Arafeh

07/05/2026

Supervised by Yuhua Li; Moderated by Richard Booth

This project investigated whether incorporating news sentiment analysis into machine learning models improve the accuracy of next day stock price direction compared to models using technical indicators. Three machine learning models from distinct architectural families were evaluated: XGBoost, LSTM, and the Transformer. Each was trained on three actively traded stocks (Apple, Tesla, and Amazon) under two feature configurations with a baseline using only technical indicators, and an extended set including sentiment scores extracted from financial news headlines using FinBERT. This produced eighteen paired experiments evaluated using accuracy, F1 score, and AUC-ROC, with statistical significance assessed through paired t-tests, Wilcoxon signed-rank tests, and Cohens’s d effect size.


Initial Plan (10/02/2026) [Zip Archive]

Final Report (07/05/2026) [Zip Archive]

Publication Form