Trend BOSS - MetaTrader 5 için gösterge

Overview

This indicator tags every bar on your chart as trending, ranging, or volatile, using a Bag-of-SFA-Symbols (BOSS) classifier built from scratch in pure MQL5. Any Expert Advisor can read that lane as context and gate its strategy on the current regime.

The method behind it is unusual, and it is the reason the indicator exists. BOSS throws the raw prices away entirely. Each window of the series is compressed into a short word of a few letters, a stretch of market becomes a bag of those words, and two stretches are compared by how similar their vocabularies are. Three differences that make raw price windows impossible to compare are each discarded by one step of that compression:

• Z-normalization removes the price level, so the same arc at 30,000 and at 60,000 becomes the same normalised curve.
• A low-pass Fourier filter removes the noise, keeping only the first few Fourier coefficients of the window and discarding the high-frequency jitter.
• Quantization into letters removes fine numerical detail, so tiny differences that do not change the shape category do not change the word.

Because the result is a bag rather than a sequence, it also does not care where inside the window a pattern happened. And because the words are literally readable, the vocabulary of a regime can be printed and inspected, which is something no elastic-distance method offers.

A single price window is z-normalized, low-pass filtered, and quantized into a short word of letters.

What Is Actually Included

The package is a complete symbolic time-series toolkit, not just the indicator on top of it. Every layer is a header-only class with no external dependency beyond ALGLIB's FFT, which ships with MetaTrader 5:

Why the Ensemble Matters

A single BOSS classifier commits to one sliding-window length, and the right length is data dependent and not known in advance. A short window captures fast local structure and misses slow sweeps; a long window does the reverse. Pick wrong and the classifier is blind to exactly the structure that separates the classes. This is a well documented weakness of the method, and it showed up plainly in testing: a single BOSS scored 25.3% macro accuracy on real bars, barely better than picking one class and calling it a day.

The ensemble fixes it. It trains one member at each of six candidate window sizes (12, 18, 24, 32, 48 and 64 bars), scores each by leave-one-out accuracy over the training bags, keeps every member that comes within 92% of the best score, and classifies by majority vote among the survivors. That single change lifted macro accuracy from 25.3% to 62.5%. The ensemble is not an optional refinement on top of BOSS; for a dictionary classifier it is what makes the method work at all.

Several BOSS members at different window sizes classify the query, and the majority vote wins.

What the Indicator Draws

The indicator trains its ensemble once when it loads, then tags each bar with the regime of the trailing segment of closes ending at it. The result is a colour-coded lane in a separate window, at three distinct heights so the states never overlap, plus an optional text label naming the current regime on the chart itself.

Reading it from an Expert Advisor. Acquire a handle with iCustom on BOSS\BOSSRegime and read buffer 0: it holds the regime code directly, 0 for range, 1 for trend, 2 for volatile, and EMPTY_VALUE on bars with no full segment behind them. Buffer 1 is only the colour index. That is the whole interface, and it is deliberately thin so a gating condition is one CopyBuffer call.

The colour-coded regime lane under a BTCUSD H1 chart: blue range, green trend, red volatile.

Recommended Setup

Why WordLen stays at 2. Keeping more Fourier coefficients means longer words that capture finer shape detail, which sounds like an improvement and is not. A coefficient sweep on a controlled two-class task showed two coefficients holding at or near 100% until the noise is heavy, and still well above chance at the extreme, while three or four coefficients collapse toward the 50% coin flip as soon as any noise appears. Fewer coefficients means a stronger low-pass filter, and a stronger low-pass filter is what survives noise.

Installation and File Structure

The headers are included as <BOSS\...> and the indicator lives in an Indicators\BOSS subfolder, so the layout below has to be preserved.

Input Parameters

Research Basis

The implementation follows the two papers that define the method:

• Patrick Schäfer, The BOSS is concerned with time series classification in the presence of noise, Data Mining and Knowledge Discovery, 2015. The bag-of-words representation, the BOSS distance, and the ensemble with its 92% retention rule.
• Patrick Schäfer and Mikael Högqvist, SFA: A Symbolic Fourier Approximation and Index for Similarity Search in High Dimensional Datasets, EDBT 2012. Symbolic Fourier Approximation and Multiple Coefficient Binning.

One point is worth stating plainly. The labels the ensemble learns from are generated by a simple statistical rule rather than by hand, so the indicator learns to reproduce that rule from window shape alone. It is a transparent, mechanical definition of regime, not an oracle, and a richer or forward-looking labeler would give BOSS a cleaner target. The engine is built so that swapping the labeler is a small change.

A companion article builds every layer from scratch, explains the mathematics, and runs the full benchmark against Dynamic Time Warping: Bag-of-SFA-Symbols (BOSS) in MQL5: Symbolic Time-Series Classification for Market Regimes.