Optimizing Short-Term Trading Strategies: Integrating Genetic Algorithms, Sentiment Analysis, and Google Trends

This dissertation examines the performance of adaptive trading strategies that apply genetic algorithms (GA), sentiment analysis, and alternative data—namely incorporating Google Trends Search Volume Index (SVI)—in optimizing market responsiveness for equity trading. Taking Apple Inc. (AAPL) during the 2023–2024 timeframe as a case study, two machine learning-based models are created and compared: an Adaptive Feature Model, in which a GA simultaneously chooses both features and model parameters, and a Fixed Feature Model, in which all engineered features are used and only model hyperparameters and risk thresholds are optimized. Both models employ an expansive feature set of technical, sentiment, and alternative data indicators. Strict train/validation/test splits are used for rigorous backtesting with performance assessed over several independent GA runs to guarantee robustness. The results show that although both adaptive models are able to capture short-term trading opportunities and gain from having flexible risk management (stop-loss and take-profit), neither of them is able to consistently beat a Buy & Hold benchmark on net return or risk-adjusted measures over the test period. The Fixed Feature Model has better robustness and generalizability, whereas the Adaptive model provides flexibility at the expense of higher overfitting risk and performance volatility. These findings highlight the continued challenge of beating good passive approaches in trending markets, the importance of careful benchmarking and open disclosure, and offer practical guidance on the construction of machine learning-based trading systems for the future.