Algorithm 1: Clustering Guided Hierarchical Activity Recognition Framework.

Input: Training set D and labels L for activities, Threshold θ, test set tx

Output: The prediction of activity with label LA of tx

// TRAINING STAGE

1. Divide D into groups CLU; // Component #1

2. calculate the activity confusion matrix CM of CLU by applying (1) and (2);

3. Construct cls_all to identify all activities; // Component #2

4. for each activity LA of L do

(4.1) S(LA) = { }; // initialization of LA

5. for each activity LA of L do

(5.1) for each activity LB of L do

obtain η(LB, LA) by applying (3);

if LA != LB and η(LB, LA) ≥ θ do

S(LA).add(LB);

(5.2) if not_empty(S(LA)) do

construct a cls_LA to distinguish LA and S(LA);

// PREDICTION STAGE

6. LA = cls_all (tx); // infer the label of tx by applying the top-level classifier

7. if not_empty(S(LA)) do

LA = cls_LA(tx); // infer the label of tx by applying the second-level classifier

8. return LA