EGCN: Entropy-based graph convolutional network for anomalous pattern detection and forecasting in real estate...

Area of Science:

• Real Estate Economics
• Data Science
• Time Series Analysis

Background:

• Real estate markets exhibit dynamic behavior influenced by economic, policy, and demographic factors.
• Traditional forecasting models struggle with market anomalies, leading to reduced accuracy and stability.
• Anomalous regions in real estate markets deviate significantly from expected trends, posing challenges for prediction.

Purpose of the Study:

• To propose a novel cluster-specific forecasting framework, EGCN, to address challenges in real estate market prediction.
• To improve forecasting accuracy and stability by independently analyzing normal and anomalous market regions.
• To enhance market insights and provide more precise, risk-adjusted predictions for real estate markets.

Main Methods:

• Developed EGCN (Enhanced Graph Convolutional Network), a framework for cluster-specific real estate forecasting.
• Implemented anomaly detection and clustering to separate anomalous regions from normal ones.
• Applied forecasting models to normal and anomalous regions independently, evaluating performance on UK, USA, and Australian datasets.

Main Results:

• EGCN achieved the lowest prediction errors compared to baseline and alternative anomaly detection methods across 12, 24, and 48-month horizons.
• EGCN demonstrated superior sensitivity in detecting anomalous regions, identifying significantly more than competing methods.
• Clustering anomalies separately reduced forecasting errors for various models, including Neural Hierarchical Interpolation for Time Series Forecasting, improving accuracy and stability.

Conclusions:

• The EGCN framework effectively enhances real estate market forecasting by treating normal and anomalous regions distinctly.
• Separately clustering anomalies allows predictive models to better capture diverse market behaviors, leading to more accurate and robust predictions.
• EGCN offers a significant advancement in real estate market analysis, providing more precise and risk-adjusted forecasting capabilities.