[Paper Review] House Price Prediction Using LSTM

Xiaochen Chen, Wei Lai+1 more|arXiv (Cornell University)|Sep 25, 2017

Housing Market and EconomicsEconomics, Econometrics and Finance2 references21 citations

TL;DR

This paper proposes a Long Short-Term Memory (LSTM) network for predicting average house prices in Beijing, Shanghai, Guangzhou, and Shenzhen for November and December 2016 using historical data from January 2004 to October 2016. The LSTM model outperforms the baseline ARIMA model in terms of Mean Squared Error, with stateful and stacked LSTM variants further improving prediction accuracy.

ABSTRACT

In this paper, we use the house price data ranging from January 2004 to October 2016 to predict the average house price of November and December in 2016 for each district in Beijing, Shanghai, Guangzhou and Shenzhen. We apply Autoregressive Integrated Moving Average model to generate the baseline while LSTM networks to build prediction model. These algorithms are compared in terms of Mean Squared Error. The result shows that the LSTM model has excellent properties with respect to predict time series. Also, stateful LSTM networks and stack LSTM networks are employed to further study the improvement of accuracy of the house prediction model.

Motivation & Objective

• To develop a deep learning model for forecasting house prices in major Chinese cities using temporal patterns in historical data.
• To compare the performance of LSTM networks against traditional time series models like ARIMA for house price prediction.
• To evaluate the impact of architectural enhancements such as stateful and stacked LSTM on prediction accuracy.
• To provide a robust, data-driven approach for real estate price forecasting in dynamic urban markets.

Proposed method

• Utilizes house price data spanning from January 2004 to October 2016 for four major Chinese cities: Beijing, Shanghai, Guangzhou, and Shenzhen.
• Employs an Autoregressive Integrated Moving Average (ARIMA) model as a baseline for comparison.
• Applies Long Short-Term Memory (LSTM) networks to model sequential dependencies in the time series data.
• Implements stateful LSTM to preserve hidden states across batches, improving temporal continuity.
• Uses stacked LSTM networks with multiple layers to enhance feature extraction and modeling capacity.
• Evaluates model performance using Mean Squared Error (MSE) as the primary metric.

Experimental results

Research questions

• RQ1 Can LSTM networks effectively model and predict house price trends in major Chinese cities using historical time series data?
• RQ2 How does the performance of LSTM compare to the traditional ARIMA model in terms of prediction accuracy?
• RQ3 To what extent do architectural modifications like stateful and stacked LSTM improve prediction accuracy?
• RQ4 Is the LSTM model robust enough to generalize across different urban real estate markets in China?

Key findings

• The LSTM model achieves significantly lower Mean Squared Error (MSE) compared to the ARIMA baseline, demonstrating superior predictive performance.
• Stateful LSTM networks improve prediction accuracy by maintaining hidden state continuity across sequences.
• Stacked LSTM networks further enhance model performance by enabling deeper learning of temporal patterns.
• The results confirm that LSTMs are well-suited for modeling complex, non-linear time series trends in real estate price data.