Deep Candlestick Mining

Abstract

A data mining process we name Deep Candlestick Mining (DCM) is developed using Randomised Decision Trees, Long Short Term Memory Recurrent Neural Networks and k-means++, and is shown to discover candlestick patterns significantly outperforming traditional ones. A test for the predictive ability of novel versus traditional candlestick patterns is devised using all significant candlestick patterns within the traditional or deep mined categories. The deep mined candlestick system demonstrates a remarkable ability to outperform the traditional system by 75.2% and 92.6% on the German Bund 10-year futures contract and EURUSD hourly data.