Study on the Prediction of Real estate Price Index based on HHGA-RBF Neural Network Algorithm

Huan Ma, Ming Chen, Jianwei ZhangInternational Journal of u- and e- Service Science and Technology 2015 · DOI

Abstract

The traditional error of the back-propagation algorithm multilayer feed-forward network (BP neural network), there are the flaws of a slow convergence of forecast, getting local minimum solutions easily, and forecast accuracy rate is not high. This paper proposes a new approach which is the combination of hierarchical genetic algorithm and least squares method to optimize the RBF neural network such that we can predict the real estate price of the Real estate Price Index. And which overcomes the shortcomings of traditional Fourier analysis, has good localized characteristics in the time domain and frequency domain, and has important value. In signal processing, image processing, voice analysis and other fields. The hierarchical genetic algorithm is usually used to optimize the topology of the RBF neural network, the radial basis function center and width. Alternatively, the least squares method could play an important role in deciding the weights of the output layer. The experimental result shows that the feasibility of RBF neural network which could be optimized by the hybrid hierarchical genetic algorithm to predict real estate closing price, and the superiority of this approach in the aspect of prediction accuracy verified in comparing with the other two methods.

Introduction

Predicting real estate price indices is a complex challenge due to the influence of various economic, political, and social factors. Traditional statistical methods, such as regression analysis or Markov models, often require large, highly regular datasets that are not always available or applicable to volatile markets. While Artificial Neural Networks (ANNs) have emerged as a powerful alternative, standard Back-Propagation (BP) networks suffer from several "Achilles' heels": slow convergence, susceptibility to local minima, and the difficulty of manually determining optimal network architecture. This study proposes a hybrid model to address these limitations by creating a self-optimizing network specifically tuned for the volatility of the Real Estate Price Index (SCI).

Methodology: The HHGA-RBF Approach

The proposed model, Hybrid Hierarchical Genetic Algorithm-Radial Basis Function (HHGA-RBF), combines the global search capabilities of genetic algorithms with the localized precision of RBF networks.

RBF Neural Network

Unlike traditional multi-layer perceptrons, RBF networks utilize a hidden layer of Gaussian functions. This architecture makes them exceptionally well-suited for local approximation and significantly faster to train than standard feed-forward networks.

Hierarchical Genetic Algorithm (HGA)

The HGA serves as the architectural designer of the network. It employs a hierarchical chromosome structure consisting of two types of genes:

  1. Control Genes (Binary): These act as switches (1 or 0) to determine the existence of hidden nodes, thereby optimizing the network's topology automatically.
  2. Parameter Genes (Real-coded): These define the center ($C_i$) and width ($\sigma_i$) of the Gaussian functions.
Hierarchical chromosome structure used for optimizing the RBF network topology and parameters.

Least Squares Method (LSM) and AIC

To enhance efficiency, the weights of the output layer are calculated mathematically using the Least Squares Method rather than being evolved by the genetic algorithm. This hybridization speeds up the learning process significantly. Furthermore, the Akaike Information Criterion (AIC) is integrated into the fitness function to penalize overly complex models, preventing overfitting and ensuring a lean, generalizable network.

Experimental Results

The study utilized Real Estate Price Index (SCI) closing prices from the Shanghai Real Estate Exchange, covering 125 trading days from July to December 2013. The model was trained on the first 107 days and tested on the remaining 18. Data was normalized to an interval of [0, 1] to ensure numerical stability.

The HHGA-RBF model was compared against a standard RBF network and a Genetic Algorithm-optimized BP network (GA-BPNN). Across multiple experimental runs, the HHGA-RBF consistently demonstrated superior performance:

  • Accuracy: The HHGA-RBF achieved a Mean Absolute Percentage Error (MAPE) of approximately 0.0076 to 0.0080, outperforming GA-BPNN (~0.009) and standard RBF (~0.012).
  • Error Metrics: The Root Mean Square Error (RMSE) for the proposed model ranged from 20.2 to 21.7, significantly lower than the competing methods (up to 29.5 for standard RBF).
Comparison of actual and predicted real estate price index values using the proposed HHGA-RBF model.

As shown in the figure above, the HHGA-RBF prediction curve tracks the actual market closing prices with high fidelity, capturing the downward trend and subsequent fluctuations more accurately than baseline models.

Conclusion

The integration of a hierarchical genetic algorithm with an RBF neural network provides a robust framework for real estate price prediction. By automating topology optimization and utilizing the least squares method for weight calculation, the HHGA-RBF model overcomes the traditional pitfalls of back-propagation networks, offering higher accuracy and faster convergence. Future work may involve testing on longer-term datasets and incorporating external economic variables like interest rates and housing policy changes to further refine predictive power.

Reference: Ma, H., Chen, M., & Zhang, J. (2015). Study on the Prediction of Real estate Price Index based on HHGA-RBF Neural Network Algorithm. International Journal of u- and e- Service, Science and Technology, 8(7), 109-118.

Summary updated

Abstract

The traditional error of the back-propagation algorithm multilayer feed-forward network (BP neural network), there are the flaws of a slow convergence of forecast, getting local minimum solutions easily, and forecast accuracy rate is not high. This paper proposes a new approach which is the combination of hierarchical genetic algorithm and least squares method to optimize the RBF neural network such that we can predict the real estate price of the Real estate Price Index. And which overcomes the shortcomings of traditional Fourier analysis, has good localized characteristics in the time domain and frequency domain, and has important value. In signal processing, image processing, voice analysis and other fields. The hierarchical genetic algorithm is usually used to optimize the topology of the RBF neural network, the radial basis function center and width. Alternatively, the least squares method could play an important role in deciding the weights of the output layer. The experimental result shows that the feasibility of RBF neural network which could be optimized by the hybrid hierarchical genetic algorithm to predict real estate closing price, and the superiority of this approach in the aspect of prediction accuracy verified in comparing with the other two methods.
BibTeX
@article{ma2015study,
  author = {Huan Ma and Ming Chen and Jianwei Zhang},
  title = {{Study on the Prediction of Real estate Price Index based on HHGA-RBF Neural Network Algorithm}},
  journal = {International Journal of u- and e- Service Science and Technology},
  year = {2015},
  doi = {10.14257/ijunesst.2015.8.7.11},
}
Figure