PRICE PREDICTIONS OF CRYPTOPUNKS USING NEURAL NETWORKS

I wanted to look at if machine learning techniques could be utilized to predict the sale prices of CryptoPunks. Using machine learning to predict stock prices is a common problem space for data scientists, setting a precedent for using such techniques to make predictions on the NFT market. I hypothesized that the rarity of a CryptoPunk determined by its type and traits would effect its price. 

​

My primary source of data for this project was a dataset from Kaggle of CryptoPunk transactions compiled from Larva Labs and Opensea from 2017-2021 which I supplemented with data from Tradingview on the daily closing price of ETH. 

​

In my analyses I was able to prove my hypothesis on the impact of punk attributes and rarity on price. Further, I fit a random forest model that predicted CryptoPunk prices with 84% accuracy. This model can be used with relative confidence to inform art collectors new to the space on the value of their investment and the price movements of NFTs.