Decoding Market Prices of Sustainable Cryptocurrencies: Fresh Insights from Ensemble Machine Learning and Explainable AI

Abstract

This research develops an integrated framework that combines ensemble machine learning and explainable artificial intelligence to predict the performance of sustainable cryptocurrencies, including Avalanche, Build and Build (BNB) Chain, Polkadot and Solana, and to reveal the dependencies between these assets and key explanatory features. The framework employs a comprehensive predictive structure that integrates supervised and unsupervised feature processing with a metaheuristic-tuned ensemble learning model. BorutaShap identifies significant explanatory variables, while isometric mapping obtains an optimized feature representation. Predictions are generated using the Extreme Gradient Boosting algorithm, with hyperparameters optimized through particle swarm optimization. To ensure interpretability, the predictive methodology undergoes rigorous analysis using multiple explainable artificial intelligence techniques that decode dependency patterns at both global and local levels, facilitating a comprehensive understanding of market dynamics for the selected assets. Results reveal that market sentiment, technological outlook and US options market fear are the primary determinants of sustainable crypto asset performance.