Machine Learning and Interval Analysis for Advanced Portfolio Optimization

Abstract

In the dynamic landscape of financial markets, portfolio optimization is a vital aspect of newlineeffective investment strategy, balancing risk and reward to achieve optimal outcomes. newlineRecent advancements in machine learning and interval analysis have introduced new newlineopportunities to enhance the precision and robustness of portfolio selection. Traditional newlineoptimization methods often fall short of capturing the complexities and volatility of newlinestock markets. By integrating machine learning, complex stock price movements can newlinebe predicted with higher accuracy, while interval analysis provides a reliable framework newlineto address uncertainties and ambiguities in financial data. newlineThis thesis presents a comprehensive approach to portfolio optimization by newlinecombining with machine learning prediction models and interval analysis to manage the newlinenonlinear dynamics of stock markets better. This approach forecasts stock prices to newlineidentify high-potential stocks, starting with machine-learning regression models including newlineRandom Forest, XGBoost, AdaBoost, SVR, KNN, and ANN. These selected stocks are newlinethen optimized using a mean Value-at-Risk (VaR) model. A sparse minimax Sharpe ratio newlinemodel is employed to optimize the performance of portfolio efficiency, with clusteringbased newlinestock selection as a precursor. Using clustering techniques such as k-means, fuzzy newlinec-means, and ward linkage, stocks are grouped by their return rates and risk levels, while newlinevalidity indices guide the selection of the optimal clusters newline