Leveraging a comprehensive set of 399 on-chain features from CryptoQuant, I used a deep learning approach (WaveNet, via GluonTS/MXNet) to forecast Bitcoin price movement for the upcoming month. This model was trained on historical data (from 2012 to present), capturing price, volume, flows, exchange metrics, miner behavior, fee stats, and more.
Key Points:
Model: WaveNet (Deep Learning, GluonTS/MXNet)
Data: 399 on-chain features (CryptoQuant), historical range: 2012–2025(including price, volume, exchange flows, miner data, network metrics, and more)
Forecast Horizon: Next 1 month
Result:
The central line shows the median predicted price.
The shaded areas illustrate uncertainty intervals (50% and 90% confidence), considering market volatility and fundamental on-chain signals.
Model expectation: Although the historical trend is down-sloping in this interval, the wide forecast intervals demonstrate the impact of macro-level uncertainty and ongoing on-chain activity.
Chart Details:
The plot below shows the median forecast with 50% and 90% confidence bands, allowing users to visually gauge both expected trend and potential volatility:
Remarks:
The model is fully data-driven and not manually optimized for market timing.
All features are sourced from CryptoQuant’s on-chain datasets, making the forecast robust to various anomalies and events in the network.
For traders and analysts, the wide forecast range may signal high-risk conditions in the coming month.
Written by CryptoOnchain
