This thesis presents an innovative method for forecasting the sequence of Foreign Exchange rates.Using Machine Learning Algorithm and especially Deep Q-learning,it was possible to extract information from Technical Indicators to spot trends in the market.We managed to considerably reduce the number of input features in the neural network compared to previous work,by using financial analysis methodology such as Technical Indicators.This allow us to have a much lighter network with similar performance.To optimize the input information,we made an analysis of the information provided by the Technical Indicators and used clustering algorithm to eliminate unnecessary information.Results show promising predictions and significant better returns over other investment strategies such as “Buy and Hold”.We were able to forecast a trend on the market: our algorithm took a short position on a decreasing market,which means that it successfully forecast a downward trend on future prices.This work has interesting implications for future studies of Market Price Predictions and is helpful to build an intelligent portfolio management system. |