STACKED DENOISING AUTOENCODER BASED PRICE PREDICTION AND CLUSTERING OF REAL ESTATE PROPERTIES
Abstract
This work develops a framework for residential real estate price prediction and clustering properties in Nova Scotia province. It differs from other studies by covering large geographic area and by defining submarkets. We also used stacked denoising autoencoder to reduce the dimensionality and used two-level clustering approach (self organizing map and K-means) for clustering real estate properties. In addition, we test with different submarkets, different training period, different geographical features,
and regularizer to improve the accuracy of the model.