Candlestick pattern detection is one of the most widely used approach to generate signals for systematic trading, especially in crypto market​. Existing pattern detection methods are mostly done by hard coded rules generated with regular statistical methods. However, the traditional approach may not generalize well to all scenarios and face challenges to incorporate additional features to enhance the prediction accuracy. In this project, we introduce a pattern detection model pipeline that detects candlestick patterns with the use of deep learning networks including NLP (Bert & Fin-Bert) and GAF-CNN. The model has the capacity to not only classify the true positive candlestick patterns more accurately, but also to easily incorporate alternative data to improve accuracy. The work is inspired by works in the field of candlestick pattern detection(Chen & Tsai, 2020; Marc Velay & Fabrice Daniel, 2018).

The main idea of the project is to convert different time series and alternative data into GAF encoded image representations and then let computer vision to extract the local features of different patterns that are hard to identify by human beings or traditional statistical methods. With use of alternative data that contains critical information about the market, the model could further classify the detected patterns into positive and negative categories that helps investors to analyze whether the specific detected pattern has higher chance of predicting future returns accurately. In addition, the model generalizes well to incorporate new datasets as it treats all inputs as an image, and therefore has good potential to be effectively incorporated into different trading strategies.


  1. Jaquart, P., Dann, D., & Weinhardt, C. (2021). Short-term bitcoin market prediction via machine learning. The Journal of Finance and Data Science, 7, 45-66. doi:10.1016/j.jfds.2021.03.001

  2. Chen, J., & Tsai, Y. (2020). Encoding candlesticks as images for pattern classification using convolutional neural networks. Financial Innovation, 6(1). doi:10.1186/s40854-020-00187-0

  3. Velay, M., & Daniel, F. (2018). Stock Chart Pattern recognition with Deep Learning. Retrieved from https://arxiv.org/pdf/1808.00418.pdf