Scholarlink

<p>The dynamic and volatile nature of foreign exchange (FOREX) markets necessitates forecasting frameworks that not only achieve statistical accuracy but also translate predictions into actionable trading strategies. Unlike most prior studies, which have predominantly focused on USD, EUR, and other major currency pairs and evaluated models using error metrics such as MSE or RMSE, this work investigates the comparatively underexplored INR-based currency pairs (EUR/INR, USD/INR, JPY/INR, GBP/INR). A novel machine learning–based decision framework is proposed, integrating statistical, machine learning, and deep learning models with engineered technical indicators—including EMA, RSI, MACD, and Bollinger Bands. Model performance is assessed through trading-oriented measures, namely Return on Investment (ROI), Sharpe Ratio, Maximum Drawdown, and Directional Accuracy, thereby bridging predictive modeling with real-world financial applicability. Empirical results reveal the superior profitability of Multi-Layer Perceptron (MLP), which achieved ROI values as high as 334.24% (GBP/INR) and 299.44% (JPY/INR), while SARIMA and Prophet exhibited strong directional accuracy (up to 86.21%). The introduction of the Swing Technical Indicators Library (STIL), a curated feature set tailored for swing trading, further enhances medium-term forecasting and trading decision support.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Based Decision Framework for FOREX Trading: Forecasting and Recommendations

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • Sejal Hanmante,
  • Shivani Patil,
  • Uzma Patil,
  • Shrikrishna Kolhar,
  • Nilkanth Deshpande,
  • Shilpa Gite

摘要

The dynamic and volatile nature of foreign exchange (FOREX) markets necessitates forecasting frameworks that not only achieve statistical accuracy but also translate predictions into actionable trading strategies. Unlike most prior studies, which have predominantly focused on USD, EUR, and other major currency pairs and evaluated models using error metrics such as MSE or RMSE, this work investigates the comparatively underexplored INR-based currency pairs (EUR/INR, USD/INR, JPY/INR, GBP/INR). A novel machine learning–based decision framework is proposed, integrating statistical, machine learning, and deep learning models with engineered technical indicators—including EMA, RSI, MACD, and Bollinger Bands. Model performance is assessed through trading-oriented measures, namely Return on Investment (ROI), Sharpe Ratio, Maximum Drawdown, and Directional Accuracy, thereby bridging predictive modeling with real-world financial applicability. Empirical results reveal the superior profitability of Multi-Layer Perceptron (MLP), which achieved ROI values as high as 334.24% (GBP/INR) and 299.44% (JPY/INR), while SARIMA and Prophet exhibited strong directional accuracy (up to 86.21%). The introduction of the Swing Technical Indicators Library (STIL), a curated feature set tailored for swing trading, further enhances medium-term forecasting and trading decision support.