Forecasting second-hand house prices in China using the GA-PSO-BP neural network model

Area of Science:

• Computational Intelligence
• Real Estate Economics
• Machine Learning

Background:

• Traditional genetic algorithms often face premature convergence, impacting the reliability of house price forecasts.
• Accurate housing price prediction is crucial for real estate investment and urban planning.

Purpose of the Study:

• To develop and evaluate a novel GA-PSO-BP neural network model for enhanced house price forecasting.
• To address the limitations of existing algorithms in predicting second-hand housing prices.

Main Methods:

• Integration of Genetic Particle Swarm Optimization (GA-PSO) with a Backpropagation (BP) neural network.
• Utilizing a dataset of 1,824 second-hand home transactions from Lianjia.com (2023-2024).
• Analysis of pivotal factors influencing housing prices in China.

Main Results:

• The GA-PSO-BP model demonstrated superior forecasting performance on complex, high-dimensional data.
• Achieved a Root Mean Square Error (RMSE) of 0.786 and a Mean Absolute Percentage Error (MAPE) of 8.9% on the test set.
• Outperformed traditional BP neural networks optimized by single algorithms.

Conclusions:

• The GA-PSO-BP neural network model significantly minimizes forecasting errors for second-hand house prices.
• Provides more accurate price forecasts in rapidly growing urban areas like Guangzhou.
• Offers valuable insights for real estate investors in dynamic markets.