Stacked Optimized Ensemble Machine Learning Model for Predicting Stock Trends through Candlestick Chart Analysis with Feature Engineering Approach

The process of predicting stock trends through the analysis of Candlestick Charts (CCs) involves interpreting the patterns formed by these candlesticks to make informed predictions about future price movements. Utilizing Machine Learning (ML) for Stock Trend Prediction (STP) through CC analysis is common in algorithmic trading. CCs provide crucial information about the high, open, closed, and low prices within a specific time rate. However, stacked ensemble methods are employed to enhance reliability and stability, which combine the predictions of multiple models. Motivated by this objective, this work introduces the Stacked Optimized Ensemble ML Techniques with a Feature Engineering Approach for STP, referred to as SOEMLT-FEA. In the training phase, various models, including Random Forests (RF), SVM (Support Vector Machine), XGBoost, Decision Tree (DT), Adaboost, and ANN (Artificial Neural Network), are trained and optimized using the Chiroptera Algorithm (CA) to fine-tune their parameters. The optimized classifiers are then ranked, and the top three models are selected as the base classifiers for a stacking ensemble method. The efficacy of the developed feature engineering approach is confirmed by the experiential outcomes obtained (2000 and 2017) in China's stock market. This approach demonstrates promising economic returns for individual portfolios and stocks, achieving a prediction accuracy exceeding 90% for specific trend patterns.