The thesis focuses on the Upward and Downward Spread Crossing Event Prediction in Cryptocurrencies using Deep Learning. The basic idea is using deep learning techniques to predict certain conditions that might happen in the limit order book, a market structure used to store user orders, which when happen can lead to arbitrage opportunities. The primary techniques used in the thesis were Transformer models and specifically the TransLOB model and also linear regression and each time we experimented with different dataset preprocessing in order to achieve better accuracy.