feature-engineering
Transform raw market data into stationary, predictive features for crypto trading models. This skill covers price, volume, technical, microstructure, and on-chain feature construction with emphasis on avoiding lookahead bias and ensuring stationarity. Feature quality matters more than model complexity in trading ML.
Feature Engineering constructs predictive signals from market data for ML trading models.
AI-generated summary based on this skill's SKILL.md
Install
agiprolabs/claude-trading-skills/feature-engineering · repository language: Python
git clone https://github.com/agiprolabs/claude-trading-skills
cp -r claude-trading-skills/skills/feature-engineering ~/.claude/skills/feature-engineeringFrequently asked questions
AI-generated answers based on this skill's SKILL.md and metadata
How do I build features for crypto trading models?
feature-engineering transforms raw market data into stationary, predictive features for crypto trading models. Start by computing technical indicators (RSI, MACD, Bollinger Bands) from OHLCV data, then layer volume and on-chain metrics like whale flows. Ensure all features are stationary—differencing or log-returns eliminate trends that cause lookahead bias. Normalize across your training window to prevent data leakage. Feature quality matters more than model complexity in trading ML.
What are the best features for trading ML models?
feature-engineering emphasizes selecting stationary, non-redundant features that capture price dynamics without lookahead bias. Prioritize technical indicators (momentum, acceleration), volume divergence signals, and microstructure features from DEX trading. On-chain features like token whale flows add predictive power. Use correlation filtering to remove redundancy, then rank by importance. Rolling statistics computed over fixed lookback windows prevent future data leakage into training.
How do I avoid lookahead bias in feature engineering?
feature-engineering prevents lookahead bias by computing all features within a fixed historical window—never using future data. Use rolling statistics with explicit lookback periods; avoid forward-filling or centering operations on price data. Normalize features per training batch, not globally. For label creation in binary classification trading, ensure labels are defined only from data available at prediction time. Stationarity checks confirm no hidden future dependence.
What on-chain features help with token price prediction?
feature-engineering covers on-chain features including whale flow tracking, transaction volume patterns, and holder concentration metrics for token trading. These microstructure signals reveal market intent before price moves. Combine with technical indicators and volume divergence to build robust feature sets. Ensure on-chain data is stationary (use differencing or log-returns) and synchronized with OHLCV timestamps to avoid temporal misalignment.
How do I normalize features for trading model training?
feature-engineering normalizes features per training batch to prevent data leakage—compute mean and std from training data only, then apply to validation/test sets. Use rolling statistics with fixed lookback windows rather than global aggregates. For price and volume features, log-returns or differencing ensure stationarity before normalization. This approach preserves temporal structure while eliminating scale artifacts that confuse ML models.
How do I select stationary, non-redundant features?
feature-engineering selects features by first ensuring stationarity through differencing, log-returns, or detrending. Then apply correlation filtering to remove redundant features that duplicate information. Rank survivors by importance using model-agnostic methods. Focus on technical indicators, volume signals, and on-chain metrics that capture distinct market regimes. Non-redundant feature sets train faster and generalize better than bloated feature matrices.
SKILL.md
rendered from the published skill — quoted content, verbatim
Feature Engineering for Trading ML
Feature engineering is the single highest-leverage activity in building ML trading models. Model selection (XGBoost vs. neural net vs. logistic regression) matters far less than the quality and diversity of input features. A simple model on great features will outperform a complex model on raw prices every time.
This skill covers constructing, validating, and selecting features from market data for use in classification (signal-classification) and regression models targeting crypto/Solana token trading.
Why Features Beat Models
Raw OHLCV data is non-stationary, noisy, and high-dimensional. Models trained directly on price series will overfit. Feature engineering transforms raw data into stationary, informative signals that capture distinct aspects of market behavior:
- Compression: Reduce thousands of price bars to dozens of descriptive statistics
- Stationarity: Convert non-stationary prices into stationary returns and ratios
- Domain knowledge: Encode
(truncated - see the full file via the links below)
Read as markdown · JSON record · Browse the source repository
File tree — 5 files
skills/feature-engineering/SKILL.md
skills/feature-engineering/references/feature_catalog.md
skills/feature-engineering/references/pitfalls.md
skills/feature-engineering/scripts/build_features.py
skills/feature-engineering/scripts/feature_importance.py