Deep learning for predicting real estate prices in Helsinki

Abstract

Accurate real estate price evaluation is beneficial for many parties involved in real estate. It is important for real estate companies, property owners, investors, banks, and financial institutes. Studies have taken several approaches to solve the problem. The latest method which is applied to the problem domain is artificial neural networks. Studies have shown promising results using artificial neural networks in real estate price evaluation and researchers have stated that neural networks are viable tool for evaluation. In this paper it is demonstrated how artificial neural networks could be applied and optimized for real estate price prediction in Helsinki, Finland. Goal is to find the most optimal model to evaluate real estate properties. Optimization of the model is achieved by researching and fine-tuning hyperparameters and model architectures, such as activation functions, optimization algorithms, loss functions, and network architectures. The models are tested on the same dataset and time interval, which enables us to create a great comparison between results. Finally, we suggest possible improvements and thoughts on ideas worth investigating.