Forex Price Prediction with Machine Learning

A machine learning pipeline for predicting forex price movements using technical indicators and market data. Built to explore quantitative finance and learn ML techniques.

Overview

This project trains multiple ML models to predict next-day returns for major currency pairs using 90+ engineered features from technical analysis, market indices, commodities, and bond yields.

Models: Random Forest, Gradient Boosting, Ridge Regression
Data: 5 years of daily forex, equity indices, commodities, bonds, VIX
Features: Technical indicators (MACD, RSI, Bollinger Bands), time features, cross-market correlations

Quick Start

# Install dependencies
pip install pandas numpy scikit-learn yfinance matplotlib seaborn scipy joblib

# Run pipeline
python fetch_data.py          # Collect data
python model.py               # Train models
python visualize_results.py   # Generate plots

Project Structure

├── fetch_data.py         # Data collection from Yahoo Finance
├── features.py           # Feature engineering (90+ indicators)
├── model.py             # Model training & evaluation
├── visualize_results.py # Performance visualization
├── models/              # Saved models & results
└── visualizations/      # Generated plots

Features

Data Collection

• 7 forex pairs (EUR/USD, GBP/USD, USD/JPY, USD/CHF, AUD/USD, USD/CAD, NZD/USD)
• 5 market indices (S&P 500, FTSE, DAX, Nikkei, ASX)
• 4 commodities (Gold, Oil, Silver, Copper)
• Bond yields (US 10Y, 2Y) and VIX

Feature Engineering

• Technical: MA, EMA, MACD, RSI, Bollinger Bands, Stochastic, ATR
• Price: Returns, log returns, momentum, ROC
• Volatility: Multi-timeframe standard deviations
• Time: Cyclical encoding (day/month)
• Cross-Market: Currency pair correlations, yield spreads

Visualization Suite

• Model comparison (RMSE, MAE, R²)
• Prediction vs actual plots
• Residual analysis with Q-Q plots
• Feature importance rankings
• Cumulative returns (ML strategy vs buy-and-hold)
• Price forecast charts

Results

EUR/USD next-day return prediction on test set:

Model	RMSE	MAE	R²
Random Forest	0.0035	0.0025	0.15
Gradient Boosting	0.0036	0.0026	0.14
Ridge Regression	0.0037	0.0027	0.12

Top Features: Recent returns, MA differences, RSI, cross-pair correlations

What I Learned

Economics & Finance: Technical analysis, macro indicators, market relationships, efficient market hypothesis

Machine Learning: Time series features, train/test splitting for sequential data, ensemble methods, cross-validation, model evaluation

Engineering: Modular Python design, data pipelines, model persistence, reproducible experiments

Example Images

Disclaimer

⚠️ Educational purposes only. Do not use for actual trading. Forex trading involves substantial risk. Past performance ≠ future results. Consult a financial advisor before investing.