Eccodax Advanced kNN Lorentziano Matrix

1. What this code is

It is a predictive indicator based on classic Machine Learning (k-Nearest Neighbors), fully implemented in PineScript v6, designed to:

Learn historical market patterns

Compare the current state with similar past states

Estimate the expected future price movement

Reconstruct a projected price consistent with the current level

It is not an oscillator, it is not a traditional technical indicator, and it does not react only to the immediate past.

2. What the Model Learns (Supervised Learning)

2.1 Features (Input Variables)

The model uses three dimensions of information, all normalized by Z-score:

Return

Measures the percentage change in price

Captures the immediate momentum of the market

Momentum (ROC)

Measures acceleration or deceleration of the movement

Differentiates trends from consolidations

Volatility

Measures the degree of market uncertainty

Adjusts the weight of strong movements vs. noise

These three variables form a market state vector.

2.2 Normalization (Z-Score)

Each feature is converted to:

Mean ≈ 0

Standard deviation ≈ 1

This ensures that:

No variable dominates the distance

The statistical comparison is valid

The model is stable in different price regimes

2.3 Target (Predicted Variable)

The model does not predict absolute price. It learns:

Observed future return after forecastBars

That is:

Learns movement, not level

Eliminates historical bias

Avoids predictions inconsistent with the current price

3. How the model makes the prediction
3.1 Search for similar patterns (k-NN)

For each current candle, the model:

Analyzes the last lookback candles

Calculates the Euclidean distance between the current state and each past state

Selects the k most similar states

Observes what happened after them

3.2 Inference

The predicted return is calculated as:

Weighted average of the future returns of the neighbors

Weights inversely proportional to the distance

More similar states → greater influence.

4. Price Reconstruction (Key Information)

From the predicted return, the model reconstructs:

Predicted Price = Current Close × (1 + Predicted Return)

Predicted Price = Current Close × (1 + Predicted Return)

This ensures that:

The forecast respects the current market level

The output is visually interpretable

There is no regression to past regimes

5. Relevant Information the Indicator Delivers

5.1 Predicted Price (Green Line)

What it is: Estimated price after forecastBars.

How to use:

Above the current price → bullish bias

Below → bearish bias

Large distance → expectation of strong movement

5.2 Predicted Return (Implicit)

Even though not plotted directly, it is the most important information in the model.

Positive → expectation of appreciation

Negative → expectation of decline

Negative → expectation of decline

Near zero → sideways market

5.3 Directional Classification (optional)

The model also acts as a binary classifier:

High if expected return > 0

Low if expected return < 0

This is used as:

Noise filter

Trend confirmation

False signal reduction

5.4 Implicit statistical context

The indicator carries information that is not visual, but is fundamental:

Market regime (trending vs. sideways)

Statistical similarity with the past

Relative confidence (via distance from neighbors)

6. What this indicator does NOT do

It is important to align expectations:

❌ Does not predict exogenous events

❌ Does not anticipate gaps

❌ Does not work well on illiquid assets

❌ Does not extrapolate long trends

k-NN replicates patterns, does not create scenarios Unprecedented.

7. Where this model works best

Markets with repetitive structure

Medium timeframes (5m – 1D)

Liquid assets

Environments with alternating regimes

8. How to use it in practice (professional recommendation)

Ideal use:

k-NN direction → bias

Technical indicator → timing

Risk management → execution

Never use it in isolation for entry.

9. Executive summary

This code delivers:

A functional supervised ML model in Pine

Prediction consistent with the current price

Statistical market direction

Reduction of historical bias

Solid foundation for quantitative strategies

Relevant information provided by this code
1. Forecasted price (line)

Statistical projection consistent with the current level

Based on similar historical patterns

2. Implicit direction

Return > 0 → bullish bias

Return < 0 → bearish bias

3. Structural robustness

Lower sensitivity to outliers

Lower scale bias

Better adaptation to different regimes

This refactored version introduces significant improvements based on modern quantitative Machine Learning practices (similar to those found in jdehorty's "Lorentzian Classification" indicator):

Lorentzian Distance: Replaces the Euclidean distance (which is affected by noise and outliers) with Lorentzian Distance, which is much more robust for financial markets.

Matrix Structure: Uses the matrix object in Pine V6 to manage training data more efficiently and cleanly than loose arrays.

Feature Engineering (WaveTrend & RSI): Replaces simple Momentum with normalized indicators (RSI, WaveTrend, CCI, ADX), better capturing market dynamics.

Min-Max Normalization: Features are normalized on a 0-100 scale so that indicators with different magnitudes do not distort the distance calculation.

Inverse Distance Weighting: Instead of a simple average, the nearest neighbors (most similar) have greater weight in the prediction.