Forecasting Bitcoin price movements persists difficult because the markets highly volatile, erratic market movements, and pronounced nonlinear behavior. These features generate significant noise in the data and make conventional predictive models less effective. To address this issue, our article introduces a hybrid forecasting framework that combines Variational Mode Decomposition “VMD” with a Bidirectional Long Short-Term Memory network “BiLSTM”. The study aims to enhance the model’s forecasting precision when dealing with large-scale nonstationary time series. The original Bitcoin price series is decomposed into several Intrinsic Mode Functions “IMFs”, allowing high-frequency noise to be separated from meaningful market dynamics. Reconstructing the signal from these modes produces a cleaner and more structured version of the data. This denoised series is then processed by a BiLSTM model, which benefits from forward and backward temporal analysis to capture richer dependencies within the sequence. The result show that VMD + BiLSTM model is more than better than others LSTM, BiLSTM, GRU, ARNN with produces RMSE = 837.47, MAE = 638.77 values and R2 = 0.9983.