Abstarct: This thesis investigates the enhancement of market forecasting by integrating artificial intelligence and economic calendars. In this study, a Forex price forecasting model for the EUR/USD currency pair is developed using the FLF-LSTM (Forex Loss Function) long short-term memory artificial neural network. The proposed model combines news analysis, technical and statistical indicators, along with raw price data. Traditional artificial intelligence methods used in market forecasting typically focus on price and chart patterns, while the economic calendar is often overlooked. This research aims to address this gap by incorporating the economic calendar, which plays a significant role in price forecasting. Previous research suggests that integrating artificial intelligence and economic calendars can significantly improve market forecasting accuracy and trend analysis. Additionally, these methods can contribute to more effective investment and risk management decisions. The final model is evaluated using MSE and R2 metrics, trained using Tanh, Sigmoid, and Relu activation functions, and presents charts of training, validation, and predicted prices. This study utilizes various datasets with different column combinations, including raw data, additional features, news headlines, news articles, and correlated columns. These additional features are incorporated into an FLF-LSTM Forex Loss Function long short-term memory model using MSE and R2 as the loss function to predict the next day's closing price. Candle drop charts for both training and validation periods, along with predicted prices, are also plotted. In conclusion, we have developed a Forex price forecaster for the EUR/USD currency pair. By incorporating news analysis and some technical and statistical indicators that Forex traders use in their daily routines, the prediction of the market's closing price can be significantly improved. Using one-day candlestick data for the EUR/USD currency pair (EURUSD), the results show that the proposed FLF-LSTM system generally reduces the Mean Absolute Error (MAE) by 10.96% compared to the classical LSTM model. Additionally, the Tanh activation function performed best on the data with generated new features.