This project investigates the application of machine learning and deep learning to short-term stock price prediction. Specifically, it conducts a controlled comparative study of four approaches — XGBoost, a one-dimensional Convolutional Neural Network, a Long Short-Term Memory network, and a novel parallel CNN-LSTM Hybrid architecture — for next-day stock return direction prediction across five stocks: Apple, Microsoft, Tesla, NVIDIA and JPMorgan Chase, covering the period 2011 to 2026. The study addresses a gap in the existing literature where inconsistent experimental conditions across studies make reliable comparison of approaches difficult. All models are evaluated under identical conditions across three experimental configurations that systematically vary lookback window size and the inclusion of news sentiment features. Sentiment signals are derived from the GDELT Global Knowledge Graph and scored using FinBERT, a BERT-based transformer model fine-tuned on financial text. A modular Python pipeline handles data collection, feature engineering, sentiment generation, model training and evaluation, with strict time-aware data splitting and multi-seed evaluation to ensure reproducible and reliable results. The primary evaluation metric is directional accuracy, measuring the proportion of test days on which the model correctly predicts whether a stock will rise or fall the following day.