Deep Learning for Bitcoin Price Prediction: The Power of Transformers and Time۲Vec

This paper presents a novel approach to Bitcoin price prediction using transformer architectures enhanced with Time۲Vec temporal embeddings, with a one-period forecast horizon. We compare three model configurations: (۱) OHLC with technical indicators, (۲) OHLC with Time۲Vec embedding, and (۳) OHLC with both technical indicators and Time۲Vec embedding. Our results demonstrate that Time۲Vec achieves comparable accuracy to traditional technical indicators while reducing input dimensionality by ۴۷%. The model shows robust performance across multiple metrics, with MSE values of ۰.۰۰۰۰۳۷-۰.۰۰۰۰۶۵ on test data, suggesting that Time۲Vec offers an efficient alternative to conventional technical indicators for cryptocurrency price prediction.

کلیدواژه ها:

Transformers ، Attention ، Time۲Vec ، Bitcoin