Machine Learning PSAR

The Machine Learning PSAR indicator is a regime-aware trend reversal system that tracks directional price movement through an adaptive Parabolic SAR, utilizing K-Means clustering for market regime detection and K-Nearest Neighbors (KNN) for signal validation. By dynamically adjusting acceleration parameters based on market context, it aims to provide more responsive signals in trending environments while reducing noise during choppy conditions.

Usage

The Usage section describes how the script can be used to identify trend reversals and quantify signal strength.

• Trend Identification: A bullish state is established when price closes above the SAR (colored blue), while a bearish state occurs when price closes below the SAR (colored red). The SAR line acts as a dynamic trailing stop.
• Confidence Filtering: Each reversal (flip) is assigned a confidence score (0–100%) based on historical similarities. High-confidence flips (typically 70%+) suggest a stronger historical precedent for success, while low-confidence flips are displayed with fainter visual fills.
• Regime Awareness: Numeric labels (1, 2, or 3) indicate the current market state. Label 3 represents a trending market (aggressive parameters), Label 2 is neutral, and Label 1 represents a choppy market (conservative parameters).
• Visual Interpretation: The opacity of the fill between the price and the SAR line reflects the KNN confidence score; denser fills indicate higher conviction.

Details

The script integrates classical technical analysis with machine learning concepts to improve the reliability of the Parabolic SAR.

• Adaptive PSAR Engine: Uses standard PSAR logic but replaces fixed increments with dynamic values scaled by market regime. It includes an optional exponential smoothing for the Acceleration Factor (AF).
• K-Means Regime Classifier: This module measures the frequency of SAR flips over a training period. It categorizes the market into one of three regimes (Choppy, Neutral, Trending) by comparing current flip frequency against historical percentiles.
• KNN Signal Validator: Upon a trend flip, the script constructs a five-dimensional feature vector (including trend duration and AF progress) and searches for the "K" most similar historical instances. The signal's confidence is derived from the outcomes of these historical neighbors.
• Kalman Smoothing Layer: A recursive Kalman filter is applied to the final output to reduce visual "stair-stepping" while maintaining zero lag during actual trend reversals.

Settings

PSAR Settings

• Base AF Start: The initial Acceleration Factor for a new trend.
• Base AF Increment: The rate at which the AF increases as the trend persists.
• Base AF Maximum: The upper limit for the Acceleration Factor.

K-Means Regime Detection

• Enable K-Means Adaptation: Toggles the dynamic adjustment of AF parameters based on market regime.
• Training Data Period: The lookback period used to classify flip frequency.
• Choppy/Trending Regime Percentile: Thresholds used to define the boundaries between market states.

KNN Signal Validation

• Enable KNN Signal Scoring: Toggles the historical validation of reversal signals.
• K - Number of Neighbors: The number of historical matches used to calculate confidence.
• Historical Lookback Period: The range of historical data scanned for similar setups.
• Minimum Confidence Filter: The threshold for classifying a flip as high-confidence.

Smoothing & Filters

• Enable AF Smoothing: Uses exponential smoothing for more fluid AF transitions.
• Enable Kalman Filter: Applies recursive smoothing to the displayed SAR dots.
• Enable Minimum Bars Filter: Requires a trend to last a set number of bars before allowing a reversal.

FAQ

How do I interpret the regime numbers on the chart? The numbers 1, 2, and 3 represent Choppy, Neutral, and Trending regimes respectively. These are determined by how frequently the market has been reversing lately; fewer reversals typically lead to a Trending (3) classification.

What is the difference between a high-confidence and low-confidence flip? A high-confidence flip occurs when the KNN model finds that similar historical price behaviors frequently led to successful trend continuations. Low-confidence flips suggest that historical outcomes were mixed or unsuccessful in similar contexts.

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