Related Experiment Videos
IALA-BPNN: superior house price prediction through multi-strategy optimized Artificial Lemming Algorithm and BP
Qifeng Guo1, Shengpeng Li2, Tianqi Xia3
1UIBE Business School, University of International Business and Economics, Beijing, 100029, China. 202374007@uibe.edu.cn.
Scientific Reports
|May 20, 2026
Summary
This study introduces an improved Artificial Lemming Algorithm (ALA) to enhance Backpropagation Neural Networks (BPNN) for accurate housing price prediction. The novel IALA-BPNN model shows superior performance in forecasting real estate trends.
Area of Science:
- Real Estate Economics
- Computational Intelligence
- Machine Learning
Background:
- Housing prices exhibit nonlinear trends, influenced by economic factors, making accurate prediction crucial.
- Backpropagation Neural Networks (BPNN) are vulnerable to initial weight and bias settings in forecasting tasks.
- Existing prediction models require enhancement to address market volatility and improve accuracy.
Purpose of the Study:
- To develop an improved Artificial Lemming Algorithm (ALA) for optimizing BPNN in housing price forecasting.
- To introduce a novel IALA-BPNN model integrating pinhole imaging learning, evolutionary mutation, and golden sine strategies.
- To validate the enhanced model's predictive accuracy and competitiveness against established methods.
Main Methods:
- Enhancement of the Artificial Lemming Algorithm (ALA) with three novel strategies: pinhole imaging learning, evolutionary mutation perturbation, and golden sine development.
- Integration of the enhanced ALA to optimize the structure and parameters of Backpropagation Neural Networks (BPNN), creating the IALA-BPNN model.
- Empirical validation using four public datasets and a specific Shanghai housing price dataset, comparing against hybrid and machine learning models.
Main Results:
- The IALA-BPNN model demonstrated significant improvements in prediction accuracy across multiple metrics compared to eight hybrid models and four machine learning models (RF, SVM, LSTM, CNN).
- Key performance enhancements include reductions in Mean Absolute Error (MAE) by 13.63% and Root Mean Squared Error (RMSE) by 14.99%.
- The model also showed notable improvements in R-squared (9.38%) and a reduction in computation time by 34.30%.
Conclusions:
- The proposed IALA-BPNN model effectively overcomes the limitations of standard BPNN for housing price forecasting.
- The integration of ALA enhancements provides a robust and competitive approach to predicting nonlinear real estate market trends.
- The model's validated performance indicates strong potential for practical application in real estate investment and policy-making.
