Deep Reinforcement Learning for Portfolio Management: An Optimized Framework for Stock Trading

Financial markets represent one of the most complex and dynamic environments where traditional prediction models consistently fail to provide sustainable competitive advantages. This study presents a novel Deep Reinforcement Learning (DRL) framework that addresses the fundamental limitations of conventional machine learning approaches in stock market prediction and portfolio management. Our research systematically evaluates seven traditional machine learning algorithms, including K-Nearest Neighbours, Decision Trees, Support Vector Machines, AdaBoost, Random Forest, Multi-Layer Perceptron, and Kolmogorov-Arnold Networks, for stock price movement classification. This evaluation demonstrates that these methods perform no better than random classifiers when applied to financial data without sophisticated preprocessing.