Comparative Analysis of Neural Network Models for REITs Portfolio Optimization using Modified K-Means Clustering and Particle Swarm Optimization

This paper proposed a comparative analysis of neural network models for Real Estate Investment Trusts (REITs) portfolio optimization, integrating dynamic optimization techniques with the Black-Litterman model to enhance predictive accuracy and decision-making. The study addresses the challenges of traditional portfolio optimization methods, which often struggle with non-linear relationships and high-dimensional data in REITs markets. The proposed methodology combines modified K-means clustering for data preprocessing, a particle swarm optimization (PSO) variant for neural network parameter tuning, and three distinct neural network architectures—Improved Backpropagation, Radial Basis Function Network (RBFN), and Convolutional Neural Network (CNN)—to predict REITs returns. These predictions are then fed into the Black-Litterman model to derive optimal portfolio weights, balancing investor views with market equilibrium. The modified K-means algorithm improves clustering robustness, while the enhanced PSO ensures efficient convergence during neural network training. Furthermore, the comparative analysis of neural networks provides insights into their respective strengths in capturing market dynamics. The experimental results demonstrate the effectiveness of the integrated approach in generating superior portfolio performance compared to conventional methods. This work contributes to the literature by offering a novel framework that bridges machine learning and financial optimization, providing practitioners with a scalable and adaptive tool for REITs portfolio management. The significance of this study lies in its potential to inform investment strategies through data-driven insights, thereby mitigating risks and maximizing returns in volatile real estate markets.