Deep Learning for Forex Trading: Leveraging Keras Sequential Models and Optimized Features to Maximize Profitability

The research presented here focuses on evaluating a predictive model's performance by leveraging diverse input features extracted from Forex price charts. A series of experi-ments explore a range of parameter settings, with assessments conducted across 30 distinct Forex symbols. The research involves utilizing OHLC (Open, High, Low, Close) data and a historical dataset of 100 candlesticks to predict price movements. It employs a trading algorithm for initiating BUY or SELL orders based on opening and closing prices, yielding suboptimal results. However, through further investigation, the study combines the MACD in-dicator and Hammer candlestick pattern, demonstrating that this integration of indicators effectively reduces noise and enhances model accuracy. It underscores the importance of continuous relearning when employing the model and recommends high-spec computer equipment. Notably, Model 3 achieved substantial profits during a 1-year and 4-month backtest.