<p>In this study, we propose a novel portfolio optimization approach combining Genetic Algorithms (GA) with Long Short-Term Memory (LSTM) networks for market predic- tion. The process starts with LSTM networks, trained on historical market data, to forecast future asset returns. These predictions guide the GA in optimizing portfolio allocations by considering multiple performance metrics, including the Sharpe ratio, Sortino ratio, and maximum drawdown. This multi-objective approach balances risk and return more ef- fectively than traditional methods, which focus on a single objective. The GA evolves portfolios through generations by selecting high-fitness candidates, performing crossover to combine strengths, and introducing mutations to explore new solutions. This iterative process leads to the most optimal portfolio allocation. Finally, the model’s performance is eval- uated using risk-adjusted returns, stability, and out-of-sample performance, with comparisons to traditional models like mean–variance optimization. Backtesting further validates the robustness and adaptability of the optimized portfolio. By integrating LSTM’s predictions with GA optimization, our approach achieves superior portfolio performance, demon- strating a more dynamic and effective solution than traditional methods.</p>