Machine Learning: kNN (New Approach)

The Machine Learning: kNN (New Approach) indicator utilizes a modified K-Nearest Neighbors algorithm to classify price data and predict future price movements based on historical proximity. This tool provides a smoothed trend-following curve and directional forecasts, offering a robust alternative to traditional moving averages and statistical models like SVM or Random Forest.

Usage

The Usage section describes how the script can be used to identify market trends and potential reversals. The indicator processes a series of price points to determine the "nearest neighbors" in a one-dimensional data space, which are then averaged to produce a prediction.

• Trend Identification: The primary output is a smooth curve that approximates the behavior of a Simple Moving Average (SMA). Users can observe the color and slope of this curve to determine the current market regime.
• Price Projection: The script provides a forward-looking "Predict" point (displayed as a circle) that suggests where the price might move in the next bar relative to the current projection base.
• Directional Hints: By analyzing the relationship between the prediction and the input data, the indicator highlights bullish or bearish shifts through color transitions.

Details

The K-Nearest Neighbors (kNN) algorithm is a non-parametric method used for classification and regression. In this implementation, the logic is adapted for time-series data:

1. Feature Construction: The script creates training records by taking consecutive price points. By default, it uses a window of 10 data points to establish the predictor values.
2. Distance Calculation: It calculates the one-dimensional Euclidean distance (absolute difference) between the current instance and historical samples.
3. Neighbor Selection: It ranks these distances to identify the "K" number of nearest neighbors based on the minimum distance.
4. Averaging: The final prediction is the average of these identified nearest neighbors.

A unique modification in this "New Approach" allows the kNN curve to behave similarly to an SMA (specifically an SMA(20) when settings are tuned to approximately N=17), but with the added benefit of machine learning-based responsiveness to local price clusters.

Settings

• Resolution: Sets the timeframe from which the data is pulled for the kNN calculation.
• # of Data Points [2:n]: Determines the window size of consecutive prices used to create training records.
• # of Nearest Neighbors (K) [1:252]: Defines the number of historical neighbors to consider when calculating the average prediction.
• Adjust Prediction: A toggle to adjust the directional logic based on the current bar versus the previous bar.
• Non-Repainting: When enabled, the indicator uses closed bar data to ensure signals do not change after the bar closes.
• Add-On: Allows the user to filter directional signals using secondary logic:
  • None: Standard kNN directional logic.
  • Pivot Point: Filters signals based on high/low pivot detection.
  • Z-Score: Filters signals based on statistical deviation from the mean.
• Pivot Point Lag / Z-Score Lag: Adjusts the lookback period for the respective Add-On filters.
• Show Outcomes: Controls the visual output (Curve only, Prediction point only, or Both).
• Projection Base: The price source (e.g., hlcc4) used as the basis for the projection.

FAQ

How does this differ from a standard Moving Average? While it looks like a moving average, the kNN approach selects data points based on their similarity (distance) in value rather than just their chronological order, allowing it to adapt more dynamically to price clusters.

What is the purpose of the "Add-On" setting? The Add-On setting provides a statistical filter (like Z-Score) or a structural filter (like Pivot Points) to confirm the kNN prediction, helping to reduce noise in sideways markets.

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