Image-based Candlestick Pattern Classification with Machine Learning

Financial markets, such as the stock market, bond market and foreign exchange market, are important channels for fund transfer. As a graphical analysis tool, candlestick charts use graphs to display the open, high, low, and close prices in a specific period. In the past, there have been attempts to identify the characteristics of candlesticks based on Gramian Angular Field (GAF) images, but they are not perfect. In this study, we implemented Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), AdaBoost, Random Forest (RF) and XGBoost models, we found that the use of deep learning models is not the best choice for the recognition of candlestick features based on GAF images. Comparing these models, MLP and CNN are better than AdaBoost and RF, but worse than XGBoost. Our results show that for the candlestick pattern classification problem based on GAF images, it is unnecessary to use complex CNNs and traditional machine learning models can also achieve satisfactory results with much less computation resources.