A Hybrid Model for Bitcoin Prices Prediction using Hidden Markov Models and Optimized LSTM Networks

With the recent advances in the Blockchain technology, and due to its decentralized nature, it has been a much considered approach for solving issues in the Internet of Things (ToT) sector, in particular, for IoT payment platforms. As Machine-to-Machine (M2M) payments are fundamental in the IoT economy, the development of Blockchain-based payment platforms, using cryptocurrency, is continuously increasing as it enables a pure M2M, secure and private financial transactions. Unlike traditional assets, cryptocurrencies have a higher index of volatility, which makes it essential to understand the movement of their prices, as a first step to optimize Blockchain-based M2M payment transactions. In this paper, we propose a novel hybrid model that deals with this challenge from a descriptive, as well as predictive points of view. We use Hidden Markov Models to describe cryptocurrencies historical movements to predict future movements with Long Short Term Memory networks. To evaluate the proposed hybrid model, we have chosen 2-minute frequency Bitcoin data from Coinbase exchange market. Our proposed model proved its effectiveness compared to traditional time-series forecasting models, ARIMA, as well as a conventional LSTM.