CurrencyMeter kNN is a multi-currency strength oscillator built for FX traders who want a clean, aggregated view of currency flows, and a simple Machine Learning layer to estimate mean-reversion probability during extreme conditions.

Unlike single-pair indicators, this tool computes strength for the 8 major currencies (USD, EUR, GBP, JPY, AUD, CAD, CHF, NZD) by aggregating information across a broad basket of FX pairs, then normalizes it into a compact meter.

How it works

1. Multi-pair aggregation
   For each currency, the meter aggregates a DMA-style signal across many pairs:

• fast MA vs slow MA (configurable periods and MA type)
• normalized as a percentage of price
  This produces an averaged “currency pressure” value per currency.

2. Extremes detection with a dynamic threshold
   Markets are non-stationary: the statistical properties of price/returns shift over time (volatility regimes, liquidity, macro cycles, session effects). A fixed threshold often fails because what is “extreme” in one regime becomes normal in another.

To reduce that instability, the model uses a dynamic threshold computed from the recent distribution of the meter (rolling history). It estimates “extreme” as the top tail of recent absolute values (percentile-based).
This transforms the raw signal into something closer to stationary behavior, which is typically easier to model and learn.

3. Simple kNN model (transparent ML)
   A lightweight k-Nearest Neighbors (kNN) classifier is used to estimate whether an extreme condition tends to revert over a defined horizon:

• Features: level + slope of the meter at the moment of the extreme
• Labeling: after ReversionHorizon bars, the model checks whether the meter moved closer to zero (reversion) or not (continuation)
• Output: a reversion probability (0–1)

When probability exceeds your Confidence Threshold, the indicator prints a directional arrow on the meter:

• Extreme low → bullish reversion expectation
• Extreme high → bearish reversion expectation

Features

• 8 major currencies plotted in one panel
• Multi-pair aggregation (reduces single-pair noise)
• Dynamic “extreme” threshold based on recent distribution
• kNN probability layer for mean-reversion bias
• Persistent on-panel arrows for recent signals
• Built-in legend for currency colors

Notes (important)

• This is intentionally a simple, transparent ML approach (kNN). It’s not a black box.
• The goal is not “prediction”, but regime-aware context: identifying when a currency becomes statistically stretched relative to its own recent behavior.

Disclaimer

This indicator is a decision-support tool, not financial advice. FX markets are probabilistic and regime-dependent, always validate settings on your broker feed and apply proper risk management.