SuperTrend AI (Clustering): The Complete Machine-Learning Trend Guide
The SuperTrend AI (Clustering) indicator, published by LuxAlgo, is one of the most downloaded machine-learning tools on TradingView — and one of the most misunderstood. It takes the classic ATR-based SuperTrend that traders have used for over a decade and wraps it in an unsupervised learning layer that continuously picks the best settings for you. No more guessing whether to use a factor of 2 or 4. The indicator runs many SuperTrends at once, scores them, and lets a k-means clustering algorithm decide which one the market is actually respecting right now.
This guide explains exactly how that works — the ordinary SuperTrend it is built on, the k-means engine that adapts it, every setting that matters, how to read the performance score, and where it fails. It is written for traders who already know what an ATR trailing stop is and want a mechanical, rules-based way to trade an adaptive one.
Automate your trades. Let Quantum Algo trade for you.
Every signal executed on your own account — 24/7, hands-off.
What the SuperTrend AI Clustering Indicator Actually Is
Start with the thing it is built on. The ordinary SuperTrend indicator plots a single line that trails price using the Average True Range (ATR). Its position is set by one number — the factor (also called the multiplier). A factor of 3 places the trailing stop three ATRs away from price. Small factors hug price and flip often; large factors sit far away and flip rarely. The eternal problem: the "right" factor changes constantly. A setting that works in a calm range gets stopped out repeatedly in a volatile trend, and vice-versa. For a related tool, see our Lorentzian Classification guide.
SuperTrend AI solves this by refusing to commit to one factor. Instead it calculates many SuperTrends simultaneously — by default from a factor of 1.0 up to 5.0 in steps of 0.5, giving nine separate trailing stops. Each one is scored on how well it has performed recently. Then a machine-learning routine groups those nine performance scores into three clusters — a best-performing group, an average group, and a worst-performing group — and the indicator adopts the ATR factor from whichever cluster you select (best, by default). As the market changes, the cluster memberships shift, and the effective factor shifts with them. The result is a SuperTrend that self-optimizes its own ATR multiplier in real time.
Multiple SuperTrends
Instead of one trailing stop, the indicator computes a spread of them across a range of ATR factors — nine by default. Each is a candidate the market may or may not be respecting.
Performance scoring
Every candidate accumulates a running performance measure — how much the market moved in its favour after each signal. This score is what the AI actually clusters on.
K-means clustering
The nine scores are grouped into three clusters (best / average / worst). The indicator trades the factor drawn from the cluster you pick, adapting as memberships change.
How K-Means Clustering Works (the AI part, in plain English)
K-means is the engine that earns this indicator its "AI" label, so it is worth understanding properly. It is an unsupervised machine-learning algorithm — meaning it finds structure in data that has no labels, no "right answers" fed in advance. According to IBM's technical overview, k-means is the most popular unsupervised clustering method precisely because it is fast, scalable, and simple: it groups unlabelled data points into a chosen number of clusters based on how close they are to each cluster's centre.
That centre is called a centroid. The algorithm works in a short loop. First it picks k starting centroids (here k = 3). Then it repeats two steps until things stop moving: assignment — every data point is attached to its nearest centroid, forming clusters; and update — each centroid moves to the mean position of the points assigned to it. Academic treatments describe the same loop: begin with a random sample of k points as centroids, assign each data point to the nearest one by distance, then recompute each centroid as the mean of its cluster, repeating until the centroids converge. The goal it quietly optimises is to minimise the total squared distance between points and their centroid — the "within-cluster sum of squares."
K-means clustering the nine SuperTrend performance scores
Nine ATR-factor SuperTrends (dots) scored on performance, grouped into three clusters. The indicator trades the factor from the cluster you choose — "Best" by default.
In SuperTrend AI the "data points" are the nine performance scores, one per ATR factor. K-means sorts them into the three clusters shown above. The best cluster holds the factors whose SuperTrends have been paying off the most; the worst cluster holds the ones getting chopped up. Because you normally set From Cluster to "Best," the indicator's live trailing stop is drawn from the top group — and it re-clusters continuously, so the chosen factor migrates as the market's character changes. One important caveat from the clustering literature: k-means results depend on the initial centroid placement, which is why the indicator lets you cap the number of iterations used to reach convergence.
The Settings That Actually Matter
SuperTrend AI exposes more inputs than a normal SuperTrend, but only a handful change behaviour meaningfully. Here is what each does and how to think about it.
| Setting | What it controls | Practical effect |
|---|---|---|
| ATR Length | Lookback for the Average True Range | Longer = smoother, slower stop. 10 is the common default. |
| Factor Range (min / max) | The span of ATR multipliers tested, e.g. 1 → 5 | Raising the minimum biases toward longer-term signals; lowering the maximum biases toward shorter-term ones. |
| Step | Increment between tested factors | Smaller step = more candidate SuperTrends = finer adaptation but heavier computation. |
| Performance Memory | How much weight recent vs. old performance carries | Higher = slower, longer-term memory; lower = reacts faster to fresh performance. |
| From Cluster | Which cluster supplies the live factor | "Best" trades the top performers; "Average" is more conservative; "Worst" is for study, not trading. |
| Maximum Iteration Steps | Cap on k-means convergence loops | Too low = poor clustering; too high = slow load. Balances quality against speed. |
Reading the Performance Score
Every signal SuperTrend AI prints carries a performance metric — a small number that quantifies how strongly the market has tended to move in the signal's direction. LuxAlgo's own documentation frames it directly: higher values suggest a stronger trend, while lower values such as 1 or 0 may indicate a retracement rather than a clean directional move. Treat it as a confidence gauge, not a guarantee.
The practical read is simple. A long signal with a high performance score is the indicator telling you the current best-cluster factor has been reliably profitable — the kind of environment where trend-following pays. A signal with a score near zero is a warning: the trailing stop may be flipping in noise. Many traders use the score as a gate — only taking signals above a threshold — which turns the raw indicator into a filtered strategy.
Three Ways to Trade SuperTrend AI
The indicator is flexible enough to anchor several styles. These three are the most robust.
Trailing-stop trend following
Enter on the trailing-stop flip, hold while the AI line stays on one side, exit on the opposite flip. The self-tuning factor keeps the stop appropriately tight or loose for current volatility.
Score-gated entries
Only take flips whose performance score clears a threshold (e.g. 5+). This filters out the low-conviction flips that occur in ranges — the single biggest source of SuperTrend whipsaw.
Confluence with structure
Use SuperTrend AI purely as a trend gate and take entries only where it aligns with order blocks, fair value gaps, or a liquidity sweep. This is where it produces the highest-edge setups.
SuperTrend AI vs. Classic SuperTrend
Is the AI version strictly better? Not always. It is more adaptive but also more complex, and complexity has costs.
| Dimension | Classic SuperTrend | SuperTrend AI (Clustering) |
|---|---|---|
| ATR factor | Fixed by you | Selected adaptively from the best cluster |
| Adapts to volatility | No — you must re-tune manually | Yes — re-clusters continuously |
| Signal confidence | None built in | Performance score on every signal |
| Computation | Light | Heavier (many SuperTrends + clustering) |
| Best for | Simple, transparent trend stops | Changing-volatility markets, filtered entries |
Where SuperTrend AI Fails
No adaptive stop escapes the fundamental weakness of all trend indicators: ranging markets. When price chops sideways, even the best-performing cluster's factor will flip back and forth, and the performance scores collapse toward zero. The AI does not make trend-following work in a range; it just tells you, through low scores, that you are in one. Ignoring that signal and taking every flip is the fastest way to lose money with this tool.
Two more honest limitations. First, k-means is sensitive to initialisation and to outliers — a single violent bar can distort the clustering, which is why the performance-memory and iteration settings exist. Second, "AI" does not mean "predictive." The indicator optimises based on past performance; a factor that clustered as "best" over the last window can underperform the moment regime changes. Adaptive is not the same as clairvoyant.
A Worked Example: How the Factor Adapts Through a Trend
Walk through what happens as a market moves from quiet to volatile. Imagine a stock consolidating in a tight range. During that phase, the low-factor SuperTrends (1.0, 1.5) sit close to price and flip constantly, racking up poor performance scores — they land in the worst cluster. The high-factor SuperTrends (4.5, 5.0) sit far from price and barely move; they neither win nor lose much. The mid-range factors around 2.5 to 3.0, which respect the range without over-reacting, tend to score best and land in the best cluster. So during consolidation the indicator quietly favours a moderate factor.
Now a breakout fires and a strong trend begins. Suddenly the tight low-factor stops start catching the move early and riding it, their performance scores climb, and they migrate into the best cluster. The indicator's effective factor tightens automatically — without you changing a single input — because the clustering has detected that aggressive stops are now the ones paying off. When the trend exhausts and price starts whipsawing again, those same tight stops get chopped, their scores fall, and the effective factor loosens back out. This continuous migration is the whole point: the trader who used a fixed factor would have been either too tight in the range or too loose in the trend, but the AI shifts between them as conditions demand.
Backtesting and Optimisation Cautions
Because SuperTrend AI is adaptive, it is tempting to assume it is immune to the curve-fitting problems that plague fixed-parameter systems. It is not. Two cautions matter. First, the Factor Range and Performance Memory are still parameters you choose — and it is entirely possible to over-optimise them to a specific historical period. A range and memory that produced beautiful equity on one asset over one window can fall apart on another. Test across multiple assets and multiple market regimes before trusting a configuration.
Second, remember what the academic literature says about k-means: its output depends on initial centroid placement and it is sensitive to outliers. A single extreme bar — a gap, a flash spike — can temporarily distort the performance scores and therefore the clustering. This is exactly why the indicator caps iteration steps and offers a performance-memory dial; both are there to keep one anomalous event from hijacking the adaptation. When you see the indicator behave strangely right after a violent candle, that is usually the clustering absorbing an outlier, and it typically normalises within a few bars. Do not re-optimise your settings in reaction to a single event.
How SuperTrend AI Fits the Wider Toolkit
The strongest use of SuperTrend AI is as a trend filter inside a structure-based system, not a standalone signal generator. Let the AI decide "are we trending, and how strongly?" while your entries come from Smart Money Concepts: take longs only when the AI is bullish with a healthy score and price is reacting from a demand order block or filling a bullish fair value gap. That confluence — adaptive trend context plus institutional price levels — is exactly the kind of setup that turns a popular retail indicator into a genuine edge.
• Order block detection — adaptive trend context meeting institutional demand and supply zones
• Liquidity sweep alerts — trend flips confirmed by a genuine stop raid, not range noise
• FVG identification — entries timed by fair value gap fills in the AI trend direction
• Multi-timeframe bias — HTF adaptive trend aligned with your lower-timeframe entries
• Smart alerts — notified when a SuperTrend AI signal meets SMC confluence
◆ Trade adaptive trends without babysitting settings
QuantumAlgo layers Smart Money Concepts — order blocks, fair value gaps, and liquidity sweeps — onto adaptive trend context, with a verified public track record so you can check every call.
See the indicator →Verify the track recordFrequently Asked Questions
It is a machine-learning trend indicator by LuxAlgo that runs many SuperTrends across a range of ATR factors, scores each on recent performance, and uses k-means clustering to group them into best, average and worst clusters. It then trades the ATR factor from the cluster you select — usually the best — so the trailing stop self-optimizes as volatility changes.
K-means is an unsupervised algorithm that groups data into k clusters around centres called centroids. Here the data points are the performance scores of each ATR-factor SuperTrend. The algorithm assigns each score to its nearest centroid, moves each centroid to the mean of its cluster, and repeats until the clusters stabilise — producing best, average and worst performance groups.
It quantifies how strongly the market has moved in a signal's direction using the active factor. Higher values suggest a stronger, more reliable trend, while values near 0 or 1 may indicate a retracement or ranging conditions. Many traders only take signals whose score clears a threshold.
It is more adaptive — it removes the need to manually pick an ATR factor and adds a confidence score — but it is heavier to compute and more complex. In simple, stable conditions a fixed SuperTrend can be just as effective. In changing-volatility markets or for filtered entries, the AI version has the advantage.
The two that matter most are Factor Range and Performance Memory. Raise the minimum factor for longer-term signals and lower the maximum for shorter-term ones; increase performance memory for slower, longer-horizon adaptation. ATR Length of 10 and the default step and iteration caps work well as a starting point.
The adaptive trailing stop itself is confirmed on bar close like a normal SuperTrend, but because the clustering re-evaluates as new data arrives, the historically selected factor can shift. Always confirm signals on closed bars and avoid acting on an in-progress bar.
It chooses which performance cluster supplies the live ATR factor. 'Best' trades the top-performing group and is the standard choice; 'Average' is more conservative; 'Worst' exists mainly for study and comparison, not for live trading.
No. It optimises based on past performance and adapts to current volatility, but it cannot predict future moves. A factor that clustered as 'best' recently can underperform the moment the market regime changes, so it should be used as an adaptive trend filter, not a forecast.
Like all trend-following stops, it flips repeatedly when price chops sideways. The difference is that the performance scores collapse toward zero in those conditions, effectively warning you. Filtering out low-score signals removes most range whipsaw.
The SuperTrend AI (Clustering) plots an adaptive trailing stop on the price chart. The AI SuperTrend Clustering Oscillator is a separate LuxAlgo tool that plots the bullish, average and bearish cluster centroids as an oscillator, with the consensus output above zero indicating an uptrend and below zero a downtrend.
It works on any timeframe, but higher timeframes (1H and above) produce cleaner clustering because there is less noise for the performance scores to react to. On very low timeframes, expect more low-score signals and use tighter score filtering.
Use it as a trend gate: take longs only when the AI trend is bullish with a healthy score and price is reacting from a demand order block or filling a bullish fair value gap, and mirror that for shorts. This pairs adaptive trend context with institutional price levels for higher-quality entries.
References & Related Guides
Primary and authoritative sources used in this guide:
- ◆ LuxAlgo — SuperTrend AI (Clustering) official indicator documentation
Definitive source on the indicator's inputs, factor range, performance memory, and cluster selection. - ◆ TradingView — SuperTrend AI (Clustering) [LuxAlgo] script & usage notes
Open-source script page describing the adaptive factor hypothesis and performance interpretation. - ◆ IBM — What is k-means clustering?
Authoritative technical overview of the unsupervised algorithm, centroids, and its known limitations. - ◆ Investopedia — Average True Range (ATR)
Background on the volatility measure that every SuperTrend factor is applied to.