IMPROVING THE AI STOCK MARKET FORECASTING WITH CANDLESTICK PATTERNS

Abstract

In the rapidly evolving digital economy, the application of Artificial Intelligence (AI) in financial forecasting has gained significant traction. This study investigates the effect of various candlestick patterns on the performance of Long Short-Term Memory (LSTM) models in predicting stock market movements. Experiments conducted on the stock price history data demonstrate that supplementing traditional input parameters (e.g., open price) with a range of candlestick patterns enhances the predictive accuracy of LSTM models. Although the initial model architecture lacked hyperparameter optimization for solving this kind of task, our findings suggest notable improvement in prediction performance when candlestick pattern flags are incorporated. Future work will focus on incorporating additional financial indicators into the model's training data and fine-tuning it through optimization algorithms to achieve greater robustness and accuracy.