Training data

Samples per activity: 60

Tree hyperparameters

Max depth: 4 Min samples / leaf: 6

Live activity stream

Stream speed: 1.0×

Live classification

True activity—

Predicted activity—

Rolling accuracy—

Samples classified0

Tree structure

Depth—

Nodes / leaves—

Training accuracy—

Every split is chosen by a real CART-style recursion: for each candidate feature and threshold it computes the Gini impurity of the two resulting groups and keeps the split with the largest impurity reduction. The top panel shows the feature-space partitions the tree has actually learned; the bottom panel draws the tree itself, each node labelled with its split rule and class counts.