NFT Price Prediction and Trait Analysis Using Machine Learning

Abstract

Non-Fungible Tokens (NFTs) represent a revolutionary class of digital assets, characterized by their uniqueness and value derived from metadata and visual traits. However, NFT markets suffer from volatility and a lack of transparent valuation systems, making it difficult for collectors and investors to estimate asset worth. This paper presents a comprehensive machine learning pipeline for predicting the market value of NFTs based on trait rarity and sale metadata. We apply rigorous preprocessing, compute rarity scores from trait distributions, and compare multiple regression models, including Random Forest, LightGBM, CatBoost, and Extra Trees. Our analysis demonstrates that tree-based models significantly outperform simpler regressors, with Extra Trees achieving the lowest RMSE of 136.91 and the highest$\mathrm{R}^{\mathrm{2}}$score of$\text{1. 0}$. Visual and statistical analyses further validate the effectiveness of our methodology in predicting NFT prices with high precision.