We explore the integration of Gaussian Processes Regression ( GPR) into portfolio optimization, presenting a novel framework that leverages predictive modeling to enhance investment performance. GPR, a non-parametric Bayesian approach, is employed to forecast asset returns and associated uncertainties, addressing key challenges in financial forecasting such as non-linearity, non-stationarity, and data noise. The study develops a Dynamic Strategy that incorporates these predictions into portfolio allocation, allowing for adaptive decision-making based on probabilistic thresholds. Results show that our dynamic strategy outperform traditional approaches such as Maximum Return, Minimum Volatility, and Maximum Sharpe Ratio in backtests across multiple metrics, demonstrating higher risk-adjusted returns and reduced transaction costs. This research advances predictive portfolio optimization, offering practical insights for risk management and decision-making while identifying future opportunities in model refinement, strategy expansion, and large-scale scalability.