Machine Learning Key Levels [AlgoAlpha]

🟠OVERVIEW
This script plots Machine Learning Key Levels on your chart by detecting historical pivot points and grouping them using agglomerative clustering to highlight price levels with the most past reactions. It combines a pivot detection, hierarchical clustering logic, and an optional silhouette method to automatically select the optimal number of key levels, giving you an adaptive way to visualize price zones where activity concentrated over time.

🟠CONCEPTS
Agglomerative clustering is a bottom-up method that starts by treating each pivot as its own cluster, then repeatedly merges the two closest clusters based on the average distance between their members until only the desired number of clusters remain. This process creates a hierarchy of groupings that can flexibly describe patterns in how price reacts around certain levels. This offers an advantage over K-means clustering, since the number of clusters does not need to be predefined. In this script, it uses an average linkage approach, where distance between clusters is computed as the average pairwise distance of all contained points.

The script finds pivot highs and lows over a set lookback period and saves them in a buffer controlled by the Pivot Memory setting. When there are at least two pivots, it groups them using agglomerative clustering: it starts with each pivot as its own group and keeps merging the closest pairs based on their average distance until the desired number of clusters is left. This number can be fixed or chosen automatically with the silhouette method, which checks how well each point fits in its cluster compared to others (higher scores mean cleaner separation). Once clustering finishes, the script takes the average price of each cluster to create key levels, sorts them, and draws horizontal lines with labels and colors showing their strength. A metrics table can also display details about the clusters to help you understand how the levels were calculated.

🟠FEATURES

1. Agglomerative clustering engine with average linkage to merge pivots into level groups.
   
2. Dynamic lines showing each cluster’s price level for clarity.
   
3. Labels indicating level strength either as percent of all pivots or raw counts.
   
4. A metrics table displaying pivot count, cluster count, silhouette score, and cluster size data.
   
5. Optional silhouette-based auto-selection of cluster count to adaptively find the best fit.
   

🟠USAGE
Add the indicator to any chart. Choose how far back to detect pivots using Pivot Length and set Pivot Memory to control how many are kept for clustering (more pivots give smoother levels but can slow performance). If you want the script to pick the number of levels automatically, enable Auto No. Levels; otherwise, set Number of Levels. The colored horizontal lines represent the calculated key levels, and circles show where pivots occurred colored by which cluster they belong to. The labels beside each level indicate its strength, so you can see which levels are supported by more pivots. If Show Metrics Table is enabled, you will see statistics about the clustering in the corner you selected. Use this tool to spot areas where price often reacts and to plan entries or exits around levels that have been significant over time. Adjust settings to better match volatility and history depth of your instrument.