Abstract
The rise of sharing economy has made the world more sustainable. Airbnb is at the forefront of this growth with more than $93 million in profit in 2017. Airbnb lists more than 500 million properties worldwide from hosts who can set their own price which could lead to either underpricing or overpricing. In this report, we approached the problem by using not only traditional listing details but also images to predict price.
To attain this, first we chose Hong Kong as a destination rich in Airbnb data. Then, we collected more than 9,000 listing details with images scraped from Airbnb website. We removed irrelevant images such as outdoors and maps. Next, we created multiple models by setting Hong Kong Dollars (HKD) price range, multiple-layer neural networks (NN) for structured data, and Convolutional Neural Networks (CNN) for images. We then designed multiple experimental neural network models when structured data and unstructured data were used.
The best results were obtained using structured data alone with a mean absolute error (MAE) of +/-167 HKD. Merging structured and unstructured data resulted to an MAE of +/-186 HKD. Both results used Airbnb listings priced from 47 HKD to 2,000 HKD. The study proved that neural networks are capable of finding complex relationships between different variables. The model can also be used to study other products that are difficult to price or value.
