Zharta Valuation of NFTs A Machine Learning Approach

Abstract

Non-Fungible Tokens (NFTs) are unique digital assets whose valuation presents a significant challenge due to their non-fungibility, low liquidity, and subjective features. This paper presents a machine learning-based approach to intra-collection NFT valuation using LightGBM, a gradient boosting model. The model was trained on historical sales, metadata, floor prices, and temporal dynamics across six prominent NFT collections. Our approach outperforms traditional valuation baselines, including floor price heuristics, rarity scores, and trait valuation models, achieving significantly lower prediction error (MAPE). The study demonstrates the potential of advanced ML models in enhancing valuation accuracy at a token-level for non-floor assets, with applications in NFT marketplace pricing, portfolio/NAV marking or NFT specialised lending.