🚀
Live ClusterForge v5 Strength pane on Bitcoin (BTCUSD) with real-time signals and performance metrics
Real-time performance table showing 84.62% win rate, benchmark comparison, and ATR expectancy
ClusterForge is a powerful custom indicator built for TradingView — the world’s most popular free platform where millions of traders view live price charts for stocks, cryptocurrencies, forex pairs, indices, and more.
On TradingView you can add custom tools called indicators that appear directly on your charts to help you analyze price action. These indicators are written in a simple programming language called PineScript.
ClusterForge v6 is one such indicator. It uses advanced machine learning (LightGBM) and behavioral clustering to detect high-probability long setups across any asset and timeframe. The signals appear in a clean lower pane with real-time performance stats so you can instantly see how well the strategy is performing.
ClusterForge started as a simple idea: one universal model that works on any asset – stocks, indices, crypto, or forex. Instead of building separate predictors for each ticker, we created a single LightGBM regression engine that learns universal price-action patterns from thousands of assets simultaneously. The result is a high-conviction, zero-drawdown signal generator that is completely timeframe-agnostic and lives directly inside TradingView via a clean Pine Script v5 pane.
Before training the final model we run a full behavioral clustering engine (pure C# KMeans with Z-score normalization). Twelve engineered features capture each asset’s true personality: AnnualizedVolatility, VolatilityOfVolatility, MomentumPersistence, VolumeSpikeFrequency, TrendStrength, MeanReversionBias, DrawdownFrequency, RecoverySpeedDays, etc. Assets with similar statistical “DNA” are grouped into coherent buckets (the “similar animals”). This clustering step dramatically improves model performance because each behavioral group can be understood and predicted in its own context.
We extract 27 stationary features (returns, RSI deviations, SMA ratios, volatility contraction, volume surges, etc.) from every bar and train one single LightGBM regressor on the entire multi-asset dataset. The model predicts the forward 5-bar % return. Signals are generated only from the top percentile of predictions, guaranteeing high-conviction setups. The latest zero-drawdown version uses the exact feature importances shown in the Pine Script edgeScore calculation.
The final output is a beautiful, self-contained Pine Script v5 indicator. It calculates the same 27 features, builds the composite edgeScore, applies percentile ranking, and adds smart filters (SMA crossover, bullish candle logic, volume surge, volatility compression, 5-bar cooldown). Every signal is backtested in real-time with win-rate, expectancy (in ATR), and a live benchmark against 500 random trades. The pane is clean, overlay-free, and ready for live alerts.