Candlestick Chart Based Stock Analysis System using Ensemble Learning

Abstract

The 1$8^{\mathrm{t}\mathrm{h}}-$century candlestick charts, which were first utilized in the Japanese rice market, are now frequently used in trading tactics across all financial markets. Candlestick charts make it possible to comprehend an asset’s opening, high, low, and closing values all in one image. In addition to these benefits, the abundance of candlestick chart patterns makes practical application challenging. A software framework that employs candlestick charts to forecast trend direction was built for this study. There are four stages of the project. A system that can identify candle patterns is developed in the first stage. The second stage involves executing training and testing procedures on data sets with labeled candlestick chart types and trend directions in order to assess the model’s performance. During the machine learning phase, open-source techniques like xgboost were applied. In the project’s final stage, it was discovered that the strategy focused solely on identifying candlestick patterns and taking positions in line with the trend based on the suggested methodology produced larger profits in 11 global indices than the Buy & Hold strategy. In comparison to the current accuracy of 53.8%, the model’s average forecast accuracy is 59.42%