Optimizing Financial Portfolios Interval Based Uncertainty Machine Learning and Metaheuristic Solutions

Abstract

In modern financial markets, optimizing investment portfolios requires sophisticated newlinestrategies that account for uncertainty, dynamic market conditions, and non-linear newlinedependencies. Traditional portfolio optimization methods often fall short in handling newlinethese complexities, necessitating the integration of advanced mathematical modeling, newlinemachine learning, and metaheuristic optimization techniques. This thesis presents newlinea comprehensive approach to portfolio optimization by incorporating interval-based newlineuncertainty representation, machine learning-driven asset selection, and metaheuristic newlinealgorithms to achieve optimal investment decisions. newlineFirst, a multi-objective optimization framework is developed, where key financial newlinemetrics such as expected return, risk, skewness, kurtosis, and entropy are modeled as newlineinterval-valued parameters. A bijective transformation and admissible order relation are newlineintroduced to handle interval uncertainty systematically, transforming the problem into a newlinedeterministic optimization model. This enables the construction of efficient portfolios newlineunder uncertain market conditions. Second, a portfolio rebalancing methodology newlineleveraging Support Vector Machines (SVM) is proposed to enhance asset selection newlinedynamically. The SVM-based classification identifies promising assets, while meanvariance newlineoptimization refines portfolio weights by balancing risk and return, incorporating newlinetransaction costs, and adapting to market fluctuations newline