Candlestick Type Prediction in Financial Charts

AI-powered financial market prediction is a leading and popular research area nowadays. While most research has focused on portfolio optimization, trend analysis, and price prediction, the primary focus of this paper is the prediction of candlestick type in financial charts. As this is a relatively new field in market prediction, few studies have been conducted, and this study aims to advance this developing topic. Candlesticks provide a visual representation of price changes over time. Candlestick patterns can be classified into reversal and continuation patterns, which can signal potential market changes. While continuation patterns indicate that the same trend continues for a bullish or bearish market, reversal patterns signal the opposite, showing transitions from upward to downward momentum or vice versa. We first proposed thirty-three attributes, including thirty-two input features and an output/prediction variable, to build our datasets. In practice, we considered four types of candlesticks—Doji, Hanging Man, Hammer, and Marubozu—that have the highest frequencies in the price chart. Second, numerous machine learning and deep learning models have been developed to achieve optimal accuracy. An ablation study validated the significant contribution of the newly engineered features, while SHAP-based explainable AI analysis highlighted their interpretability and impact on model decisions. When evaluated on hourly data, the proposed Hybrid Transformer-LSTM (HTL) model could enhance performance, achieving an accuracy of approximately 86%.