The decentralized nature of Bitcoin, Ethereum, and Dogecoin wrangles with conventional financial forecasting techniques. Because cryptocurrencies move in unpredictable patterns, forecasting their prices remains a difficult task. The study determines how Informer stacks up against LSTM networks, Random Forest, CNN and Prophet models when predicting cryptocurrency prices while operating in volatile markets. The study showed that a combination of Informer model’s ProbSparse self-attention mechanism and generative-style decoding framework enabled better accuracy along with scalable capabilities when handling long-sequence forecasts. The LSTM networks produced good results for sequential pattern recognition yet they encountered limitations regarding computational processing expenses as well as overfitting issues. The short trends were identified efficiently by CNNs while the model found it difficult to handle long dependencies across sequences. Random Forest delivered exceptional results in steady markets although it failed to grasp time-related patterns and Prophet demonstrated poor performance in unpredictable environments. The findings of this research indicate that the Informer model demonstrates effective capacities in handling unpredictable market characteristics and indicates potential as a leading prediction tool. The evaluation of Bitcoin price predictions shows Informer delivers a better forecasting outcome through its MAE = 3243.27 and R2 value of 0.80 compared to LSTM (MAE = 3246.05, R2 = 0.68) and CNN (MAE = 18981.60, R2 = −1.04). In volatile market conditions Informer achieved the best results for both Ethereum and Dogecoin with MAE = 22258.91 and 0.0127 and R2 scores = 0.91 and 0.9845, thus proving its superior performance.