Pushing the boundaries of the efficient market hypothesis, this thesis explores how well instances of multi-layered sequential long short-term memory (MLS-LSTM) recurrent neural networks augmented with Fama-French factors generate alpha—overperformance compared to a buy-and-hold strategy. Testing 30 assets including large-cap stocks, small-caps, and cryptocurrencies, this study finds that incorporating Fama-French factors boosts alpha generation by an average of 45.26 percentage points with statistical significance. Models demonstrate their highest alpha with cryptocurrencies, consistent with their lower informational efficiency, but statistical significance for the effects of market segmentation varies by analysis method. Robust analytical techniques including ANOVA, multiple linear regression, and Sharpe ratios are used to validate results. Findings affirm that incorporating broad market information can improve machine learning performance, while limitations include sample size, timeframe, and computational constraints. Enriched, well-designed neural networks have a high potential to exploit market inefficiencies, particularly with volatile low-information assets like cryptocurrencies.