ml-strategy

Generate machine learning trading signals from OHLCV data by predicting the sign of future returns with walk-forward training, converting predicted probabilities into continuous signals in the [-1.0, 1.0] range.

Walk-forward validation prevents data leakage in backtesting pipelines by training an sklearn classifier using expanding or sliding windows, ensuring feature engineering and label alignment remain strictly chronological across time.

Yes, you can use scikit-learn to predict the sign of future returns for a portfolio of symbols. The skill applies walk-forward training to compute daily long or short strength signals using a 5-day prediction horizon.

Feature engineering from market data derives momentum, volatility, RSI, moving-average ratios, volume ratios, Bollinger Band position, and intraday ratios, applying division-by-zero guards and inf or NaN sanitization for leakage safety.

OHLCV validation handles missing data by checking required columns, minimum data length, and NaN ratio, automatically skipping any symbols that fail these quality gates before feature construction begins.

Limitations of using sklearn classifiers for continuous signal mapping include the requirement for sanitized feature construction and NaN or inf handling, as the model outputs probabilities that must be safely mapped to the [-1.0, 1.0] range.