This paper addresses a novel approach to cryptocurrency risk management. An adaptive fuzzy model based on level sets is suggested to model and forecast the realized volatility of Bitcoin. The model, referred to as adaptive level set model (ALSM), is a rule-based fuzzy inference system that uses the concept of level sets to determine the model output in a data-driven and adaptive manner. One-step-ahead forecasts of realized volatility generated by the ALSM model are evaluated in terms of accuracy and compared to alternative machine learning models, an evolving fuzzy model, and the heterogeneous autoregressive (HAR) model. HAR serves as the baseline for realized volatility forecasting evaluation. The results indicate that the ALSM model achieves the highest accuracy among all competing approaches, highlighting its potential as a valuable tool to assist investors in forecasting risk in the Bitcoin market.