Intelligent price alert system for digital assets - cryptocurrencies

Abstract

Cryptocurrency market is very volatile, trading prices for some tokens can experience a sudden spike up or downturn in a matter of minutes. As a result, traders are facing difficulty following with all the trading price movements unless they are monitoring them manually. Hence, we propose a real-time alert system for monitoring those trading prices, sending notifications to users if any target prices match or an anomaly occurs. We adopt a streaming platform as the backbone of our system. It can handle thousands of messages per second with low latency rate at an average of 19 seconds on our testing environment. Long-Short-Term-Memory (LSTM) model is used as an anomaly detector. We compare the impact of five different data normalisation approaches with LSTM model on Bitcoin price dataset. The result shows that decimal scaling produces only Mean Absolute Percentage Error (MAPE) of 8.4 per cent prediction error rate on daily price data, which is the best performance achieved compared to other observed methods. However, with one-minute price dataset, our model produces higher prediction error making it impractical to distinguish between normal and anomaly points of price movement.