SuperTrend AI Clustering runs nine SuperTrends at once, scores each one on how it would
have performed, and draws the average of whichever group is winning. On MT4 and MT5. The
factor you never wanted to optimise gets picked for you, and it changes as the market does.
The number nobody knows how to set
A SuperTrend has one setting that matters: the ATR multiplier. Set it at 1 and you get
whipsawed. Set it at 5 and you give back half of every move. Every forum thread about
SuperTrend is really a thread about that number, and the honest answer is that the right
value is different on gold than on EURUSD, and different this month than last.
So this doesn't ask you. It runs the whole range.
Factor Range min 1 to Factor Range max 5 in steps of 0.5 gives nine SuperTrends running
in parallel on every bar. Each one carries a running performance score: an exponential
average of what it would have made, with Performance Memory (10) setting how far back it
remembers.
The clustering part
Nine scores, three groups. The centroids start at the 25th, 50th and 75th percentile of the
scores, then k-means moves them until they settle: best performers, middling, worst.
From Cluster picks which group to trade, and the default is best. The factor actually
used is the average of the factors that landed in that group, so it's not "the single best
setting", which would jump around. It's the centre of the group that's working.
Set it to worst if you want to see what the argument looks like from the other side. Most
people won't.
What you get on the chart
The heavy line is the trailing stop: teal below price in an uptrend, red above it in a
downtrend, stepping as the factor and ATR move. Arrows mark the flips.
Behind it runs a fainter line, the adaptive average, which moves at a rate tied to the
performance index. When the chosen cluster is doing well it tracks closely; when it isn't,
it drifts. It's a second opinion on the same signal rather than a signal of its own.
Candle Coloring tints the candles to the current direction and is on by default.
Show Dashboard gives you the performance readout.
MT4 and MT5
Same nine factors, same scoring, same clustering, same output. Below is the MT4 build.
The two builds sit at different version numbers, 1.04 on MT5 and 1.02 on MT4, because the
MT5 side has had two rounds of fixes to its buffer handling that never applied to MT4. The
calculation is the same on both.
The cost, and I'd rather say it plainly
Nine SuperTrends and a k-means loop per bar is not free. Historical Bars Calculation caps
the work at 10000 bars and Maximum Iteration Steps at 1000, and those caps are there
because without them the first load on a full chart takes a long time.
On a fresh chart expect a pause before anything appears. On M1 with a long history, expect a
longer one. This is the most expensive indicator in the set to run and there's no way to
make adaptive-anything cheap.
Repainting, honestly
The calculation runs when a bar opens, not on ticks, and the trailing stop is a ratchet: in
an uptrend it only moves up, in a downtrend only down. Past values don't move.
Here's the part that needs saying, though. The chosen factor is picked from performance
scores computed over the bars up to that point, and those scores change as new bars arrive.
The current factor can therefore be different tomorrow from what it is today, even though
yesterday's drawn line stays put. So the history you're looking at was drawn with the
factor that was being chosen at the time, which is the right way round, but it does mean the
chart is not showing you one fixed configuration end to end.
That's inherent to any adaptive method. It's worth understanding before you judge the tool
by how good the old part of the chart looks.
What it doesn't do
- No entries, stops or targets. The trailing stop is a level; what you do at a flip is
yours. - No multi-timeframe.
- Notifications default to none. Turn them on if you want the flips announced.
- No way to see the nine individual SuperTrends. You get the chosen one and the
dashboard's summary, not the whole spread. - The performance score is a simple running average of bar-to-bar moves, not a backtest
with costs. It has no idea about spread or slippage.
Settings I'd start with
H1 or H4, defaults, From Cluster on best. The point of the tool is that you don't tune it,
and if you find yourself tuning the factor range you've bought the wrong indicator.
Performance Memory at 10 is the one number I'd consider changing. Higher makes the choice
of cluster more stable and slower to react to a regime change; lower does the opposite.
Leave ATR Length at 10.
Questions you'll have
Does SuperTrend AI Clustering repaint? The drawn line doesn't. It updates on bar open
and the trailing stop only moves one way. But the factor it selects is chosen from
performance scores that keep updating, so the setting in use today is not necessarily the
one in use next week. That's what adaptive means.
Which ATR multiplier is it actually using? The average of the factors in the chosen
cluster, recomputed each bar. The dashboard shows the performance side of it.
What does From Cluster do? Picks whether to follow the best, middling or worst
performing group of factors. Best is the default.
Why is it slow to load? Nine SuperTrends plus a k-means loop on every bar, capped at
10000 bars. The pause on first load is that work.
Does it work on M5? It runs. The flips get frequent and the performance scores get
noisy, and I'd rather use it on H1 and above.
Why are the MT4 and MT5 versions numbered differently? The MT5 build has had two buffer
fixes that didn't apply to MT4. The calculation is identical.
Are there alerts? Yes, but notifications default to off.
MT4? Both builds are included, one download each.
Both platform builds come with a subscription. No win rate appears in this listing, and the
performance score inside the indicator is not one either: it's a relative ranking used to
choose a setting, computed without spread or costs, and it is not a claim about what the
tool earns.