Machine Learning BBPct

The Machine Learning BBPct indicator provides an enhanced Bollinger Band %B oscillator integrated with a simplified K-Nearest Neighbors (KNN) pattern matcher to reduce noise and provide context-aware market state descriptions. By comparing current volatility, momentum, and volume features to historical analogs, the tool blends past consensus with raw %B values to improve signal clarity.

Usage

The Usage section focuses on interpreting the hybrid oscillator and its adaptive features for market analysis. Users can utilize the script to identify overbought/oversold conditions while benefiting from an ML layer that attempts to filter out "whippy" price action common in fast-moving regimes.

• Mean Reversion: In range-bound markets, use the ML-smoothed %B and adaptive thresholds to identify potential reversal points with fewer false signals than standard %B.
• Trend Analysis: During persistent trends, observe how the KNN blend maintains the signal line within extreme zones. A high ML confidence score suggests historical support for the current price positioning.
• Contextual Filtering: The "ML Prediction" dots provide a secondary visual guide. When these dots turn bright white, it indicates high model confidence (>0.7), suggesting the current market state strongly matches historical patterns.

Details

The script operates through a multi-step logic process that begins with standard Bollinger Band %B calculations.

• Feature Engineering: The KNN model constructs a "feature vector" for each bar using normalized values for %B, band width (volatility), price momentum (ROC), volume momentum, and price position within the bands.
• KNN Logic: The script uses Euclidean distance to rank the most similar recent bars within a specified historical window. It calculates an inverse-distance weighted average of the neighbors' prior %B values to generate a prediction.
• Adaptive Filtering: The "Adaptive" filter uses the model's confidence to modulate smoothing. High confidence leads to stronger reliance on the ML estimate, while low confidence reduces its influence.
• Regime Awareness: The tool calculates a volatility regime (ATR relative to its SMA) to nudge the overbought/oversold thresholds. This ensures that extreme readings are comparable across high and low volatility environments.

Settings

BBPct Settings

• Price Source: Determines the data point (e.g., Close) used for %B calculation.
• Calculation Period: Sets the lookback for the SMA basis and standard deviation.
• Multiplier: Adjusts the standard deviation width of the bands.
• Smoothing (Toggle/Type/Length): Allows for pre-processing the %B stream using various moving averages like EMA, HMA, or TEMA.

KNN Machine Learning

• Enable KNN: Toggles the machine learning layer on or off.
• K (Neighbors): The number of historical analogs used to generate the prediction.
• Historical Period: The size of the search window for finding neighbors.
• ML Weight: Controls the blend between raw %B (0.0) and the KNN estimate (1.0).
• Number of Features: Selects between 2 to 5 data points to include in the pattern matching.

Filtering & Optimization

• Filter Method: Selects between None, Adaptive, Kalman, or Hull smoothing for the final output.
• Enable Self-Optimization: Enables a rolling comparison of parameter sets to find the most stable K value.
• Adaptive Thresholds: Automatically adjusts OB/OS lines based on current volatility levels.

FAQ

What do the signal labels like "Early Buy" mean? These are heuristic markers based on a short-term slope and acceleration probe. They indicate that the oscillator is on track to cross a threshold soon, provided momentum remains consistent.

How is the "Win Rate" calculated? The script includes an informational tracker that scores past signal outcomes based on simple target, stop, and time-out logic. This is a diagnostic tool and not a robust backtest.

How can I access this indicator? You can get access on the LuxAlgo Library for charting platforms like TradingView, MetaTrader (MT4/MT5), and NinjaTrader for free.