Dynamic Mutation-Driven Random Forest for Robust Algorithmic Trading in Volatile Markets

Algorithmic trading in volatile financial markets presents significant challenges due to abrupt price fluctuations, regime shifts, and noise-driven anomalies. Conventional machine learning models often fail to adapt dynamically, resulting in reduced profitability and heightened risk exposure. Prior studies leveraging Random Forests, Support Vector Machines, and Deep Neural Networks demonstrate predictive capabilities but suffer from overfitting, static parameterization, and poor generalization in rapidly changing market conditions. Moreover, existing ensemble models lack adaptive mutation mechanisms to recalibrate under stress scenarios. We introduce Dynamic Mutation-Driven Random Forest (DMRF), a novel ensemble learning framework that integrates evolutionary-inspired mutation strategies into tree construction. By dynamically mutating feature subsets and split thresholds during training, DMRF enhances diversity, reduces bias, and increases robustness against volatility-induced noise. Experiments were conducted on high-frequency intraday trading data from the S&P 500 index (2015–2023), encompassing 1-minute OHLCV (Open, High, Low, Close, Volume) records alongside derived technical indicators. Data normalization, volatility clustering detection, rolling-window feature engineering, and noise reduction using wavelet transforms were applied to ensure stable learning signals. DMRF achieved superior predictive accuracy (94.7%), F1-score (0.92), and Sharpe ratio (2.14) compared to baseline Random Forests and LSTM-based models. Notably, drawdown risk was reduced by 18.6%, underscoring the model’s resilience in highly volatile conditions. This work contributes a mutation-driven adaptive ensemble paradigm tailored for financial markets, demonstrating improved predictive stability, profitability, and risk management. The proposed framework advances robust algorithmic trading strategies under uncertainty. DMRF lays the foundation for adaptive, mutation-augmented machine learning architectures in finance, with potential extensions to portfolio optimization and real-time risk assessment.