A Machine Learning Approach in Predicting NFT Prices

Abstract

We construct a Non-Fungible Tokens (NFT) dataset to develop a forecasting pricing model. Forecasting the price of NFT is especially hard because NFTs have a wide diversity of properties and utilities, from acting as a membership card to getting “drops” of additional NFTs/tokens, to access to selective Discord channels, different branding/blockchain hosting NFT/market place/activation of specific abilities in games, etc. making it difficult to get fair comparables. However, because NFT trading happens on public blockchains and enough time has passed since the introduction of NFTs, a plethora of data exists suitable for training Machine Learning (ML) models. This work advances over previous work by factoring the utility of the NFT as features and by utilizing this now larger dataset for the model training. Advanced deep-learning ML models are trained, benchmarked against a holdout test set, and compared to commercially available ML models. This work provides better predictability of NFT pricing and thus opens the asset class to a broader set of investors.

Date of Award	Aug 2023
Original language	American English
Supervisor	Davor Svetinovic (Supervisor)