A1743
Title: Using images as covariates: Measuring curb appeal with deep learning
Authors: Matt Webb - Carleton University (Canada) [presenting]
Abstract: The purpose is to detail an innovative methodology to integrate image data into traditional econometric models. Motivated by forecasting sales prices for residential real estate, the power of deep learning is harnessed to add "information" contained in images as covariates. Specifically, image features are extracted using the ResNet-50 architecture and subsequently compressed using autoencoders. Forecasts from a neural network trained on the encoded data result in out-of-sample predictive power. These image-based forecasts are also combined with standard hedonic real estate data, resulting in a unified dataset. It is shown that image-based forecasts increase the accuracy of forecasts when regarded as an additional covariate. It is also the attempt to explain which covariates the image-based forecasts are most highly correlated with. The benefits of interdisciplinary methodologies are exemplified by merging machine learning and econometrics to harness untapped data sources for more accurate forecasting.
