A Comparative Analysis of Machine Learning Algorithms for Stock Portfolio Optimization

Portfolio diversification is crucial for mitigating investment risks, yet many investors inadvertently concentrate their holdings in similar types of stocks, increasing their overall risk exposure. In this study, we propose a multi-layered machine learning approach to address this challenge. Our methodology involves three main tasks: (1) fitting stocks using historical data and regression models to create effective representations, (2) clustering stocks based on these representations to identify similarity groups, and (3) utilizing the clustering model to assist investors in diversifying their portfolios. We employ various regression and classification techniques, including linear regression, support vector regression, decision tree regression, and KNN classification, along with clustering algorithms such as K-Means. Our results indicate that employing a linear regression model with standardization yields the best performance for stock representation. By leveraging machine learning techniques, we provide investors with actionable insights to enhance portfolio diversification and mitigate risks effectively.