The Unsupervised Clustering Strategy is a machine learning approach that segments forex market conditions into distinct regimes or behaviour patterns without predefined labels. By grouping similar data points (e.g. price action, volatility, momentum) into clusters, this strategy enables traders to tailor trade decisions based on the current market type, such as trending, ranging, or high-volatility phases.

Ideal for quant traders, AI developers, and system designers, this strategy is especially powerful in environments where market regimes shift often and traditional indicators underperform.

What Is Unsupervised Clustering in Trading?

Unsupervised clustering uses algorithms to identify natural groupings within data. In forex trading, clustering helps classify:

• Market states (e.g. trend vs chop)
• Volatility zones (low, medium, high)
• Pattern frequency or candlestick behaviour
• Asset correlations and co-movement

Popular clustering algorithms:

• K-Means – partitions data into k groups by minimising intra-cluster variance
• DBSCAN – detects dense clusters and outliers
• Hierarchical clustering – builds a tree of nested clusters
• Gaussian Mixture Models (GMM) – probabilistic clustering based on distribution assumptions

Strategy Workflow

1. Feature Engineering

Collect and prepare features from historical price data:

• Price-based: returns, log returns, rolling averages
• Volatility: ATR, standard deviation, Bollinger Band width
• Momentum: RSI, MACD histogram, rate of change
• Volume (if available)
• Optional: time of day, session, macro indicator shocks

Normalise or standardise all inputs for clustering.

2. Cluster Formation

Choose a clustering algorithm (e.g. K-Means with 3 clusters):

• Fit the model on training data
• Assign each data point (candle or window) to a cluster
• Interpret clusters by backtesting the performance of simple strategies in each regime

Example cluster labels:

• Cluster 0 = low volatility, choppy
• Cluster 1 = strong trend, bullish
• Cluster 2 = high volatility, reversal-prone

3. Cluster-Based Trading Logic

Create a unique strategy for each cluster:

Cluster 0 (chop):

• Use mean-reversion trades with Bollinger Band reversion
• Take profit at moving average centre

Cluster 1 (trend):

• Use breakout strategy with MA filter
• Ride trend with trailing stop

Cluster 2 (volatile):

• Reduce position size
• Trade only with tight stops or fade extreme spikes

Real-Time Application

• Use a sliding window of recent candles to generate features
• Predict cluster label in real time
• Route trade signals to the appropriate rule set
• Adjust risk parameters and position sizing based on detected regime

Example: EUR/USD Clustering Strategy

• Features: 5-period return, 20-period ATR, RSI, BB width
• Algorithm: K-Means with 3 clusters
• Cluster definitions (based on backtesting):
  • Cluster 0: sideways — use RSI reversals
  • Cluster 1: bullish trend — use breakout entries
  • Cluster 2: volatility spike — stay out or fade failed breakouts

Backtest results with adaptive logic:

• Sharpe ratio: 1.88
• Win rate: 63%
• Max drawdown: 6.2%
• Return: +21% annually

Tools and Libraries

• Python: scikit-learn, pandas, NumPy, Backtrader
• Visualization: Seaborn, Matplotlib (e.g. PCA plots for clusters)
• Deployment: MetaTrader, cTrader via Python bridge
• Real-time updates: Use rolling window + batch inference

Advantages

• Adapts to changing market regimes dynamically
• Removes need for manual state labelling
• Works well with both rule-based and AI-driven strategies
• Improves signal quality by isolating poor environments
• Easy to combine with supervised learning or reinforcement models

Limitations

• Cluster definitions can shift over time
• Requires retraining to remain effective
• Not interpretable by default — cluster meaning must be deduced
• Computational cost increases with more features and frequent updates

Best Currency Pairs

• EUR/USD, GBP/USD, USD/JPY – exhibit clear regime shifts
• AUD/JPY, GBP/JPY – highly reactive to volatility and trend states

Conclusion

The Unsupervised Clustering Strategy offers a powerful way to understand and trade market conditions by detecting natural structure in the data. By segmenting price behaviour into clusters, traders can execute strategies that are tailored to each regime, dramatically improving trade timing and performance.

To learn how to build clustering models, engineer regime features, and integrate AI-based strategies into real-time FX systems, enrol in the advanced Trading Courses at Traders MBA.