Candlestick Patterns Recognition in Bitcoin Price Action Graph Chart Using Deep Learning

Abstract

Deep learning techniques have attained great success in several computer vision and object recognition applications. However, the field of stocks, market trading prediction, and analysis using computer vision and deep learning still faces a big challenge and presents a less state-of-the-art contribution. Cryptocurrency trading is the operation of buying and selling assets through a cryptocurrency exchange platform or a trading platform in general. In this chapter, we used a simple and low-cost Convolutional Neural Network architecture to encode and recognize powerful candlestick patterns in the graph chart of Bitcoin cryptocurrency in order to predict its potential future activity. Candlestick presents a strong visual indicator that can provide important information about price action. Therefore, using deep learning techniques to identify these patterns can automate decisions in the market and preserve the trader&s;s discipline. In addition to that, the recognition of candlestick patterns in historical data allows traders to perform Backtesting and test several strategies. This will help traders manage multiple trades and make accurate decisions. Two different forms of data were implemented to compare the prediction results. The first dataset is one-dimensional based on OHLC (Open, High, Low, Close) Bitcoin price data, and the second one is based on images of two powerful candlesticks on the Bitcoin trading chart. Our approach was trained using Bitcoin historical data and performs perfectly on test data with an accuracy of 99%.