利用google街景圖片估計周遭環境對房價影響

Abstract

由於 Google 街景在視覺上描繪了具有不同社會特徵的區域，因此我們能夠藉由卷積神經網絡模型來分析這些街景環境，並分析其對房價的影響。利用此模型則不需手動分類和判斷，而是分解各個街景圖的像素後，給予每一張 Google 街景一個潛在的對應分數。再把這個分數增加至享樂住房定價模型中，我們發現其具有統計意義且能夠增加了模型的決定係數。實驗顯示，Google 街景確實能夠提供有關住宅位置的視覺線索並改進了傳統的享樂模型。此外，我們使用特徵模型和不同的機器學習模型來預測房價，均降低了均方根預測誤差，且此得分特徵具有相對較強的預測能力。這些結果表明，通過使用先進的電腦視覺技術，可以提高模型的可解釋性和預測精度。

As Google Street Views visually depict areas with disparate social characteristics, we use them to analyse the effects of environmentally locational factors on housing prices by constructing a convolutional neural network model. Instead of manual classification and judgement, the model decomposes a view's pixels then assigns a latent score for each Google Street View corresponding to the housing price. Putting this score into a hedonic housing pricing model, we find that this score has a statistical significance and increases the coefficient of determination of the model. It is empirically shown that Google Street Views provide visual cues regarding the dwelling's location and improve the traditional hedonic model. By comparing the importance of all variable in the hedonic model, this score factor takes the second position only behind the building area factor. In addition, we use hedonic models and machine learning models to predict housing prices, it reduces the root mean square prediction error and the score factor has relatively stronger predictive power. These results suggest that by using advanced computer vision technology, it can enhance the model interpretability and prediction accuracy.