When Data Science Meets NFT

TL;DR

I created an open dataset for the Crypto Coven NFT collection using OpenSea API and my scripts (it took about 4 hours to download all images), which can be used to teach/practice various topics, such as exploratory data analysis, time series analysis, data visualization, predictive analytics, text mining, and image analytics via deep learning.

The dataset can be accessed at Kaggle. I also added some simple code notebooks and look forward to more contributions from the community:

• NFT Sales Price Prediction using AutoML
• Witch Image Similarity using Tensorflow and Scikit-learn

Analysis Examples

I have used this dataset for assignments in the undergraduate data analytics and machine learning courses I teach.

I list some analysis examples here.

• What’s the percentage of witches that have not been sold at all?

  50%

• Find the top 10 addresses holding most witches. Based on the average sales price, how much those witches are worth?

  This address 0xa1707c82aa2866955991c7f2c6f431d6619b8b4c owns 162 witches, which are worth ~ $595,801 based on the average sales price in the past. Check out this address on OpenSea: https://opensea.io/Divine_Femininity

• Which is the most expensive witch ever sold (before 4/21/2022)? Who bought it?

  It’s her, bought for $39,772 on January 25, 2022 by https://opensea.io/kaiynne2.

• Plot the daily sales price and weekly moving average to understand the NFT investment cycles.

  

• Find witches who look alike

  Witch image features are extracted using pre-trained Xception model and similar images are calculated using cosine similarity and the top N nearest neighbors are returned.

  See some results (I did not know there are smoker witches ^_^ before this analysis):

  

I hope you find this dataset interesting and useful and use it in your teaching and learning.

PS. I made the witch collage shown at the top of this post using the images in this dataset, which can be downloaded from Figma.com.