Algorithm 1. Diversified dynamic class weighted ensemble.

Procedure: DDCW({X, Y}, p, k, l, α, β, θ)

Input: Data and labels {x, y}, chunk size p, min_experts k, max_experts l, fading factor α, multiplier β, threshold θ

Output: Global predictions G

1: Experts ← create_random_experts(k);

2: initialize class weights wi,j

3: for s = 0,…,n do

4: for i = 1,…, num_experts(Experts) do

5: Local_predictions = classify(Expertsi, xs);

6: if Local_predictions = ys then

7: wi,L = β * wi,L; ← Multiply weight of particular expert and target class from local prediction by β

8: end if

9: end for

10: if all samples in a chunk are processed then

11: Local_predictions = classify(Experts, x_s);

12: Diversity = calculate_diversity(Local_predictions, y_s);

13: for i = 1,…, num_experts(Experts) do

14: expert_lifetime ← Increase expert lifetime in each period;

15: wi = wi – (exp(α * expert_lifetime) – 1)/10;

16: wi = wi (1 – Diversityi);

17: end for

18: end if

19: for j = 0,…,Class_labels do

20: Global_predictionsj ← sum(wj);

21: end for

22: Global_predictions ← argmax(Global_predictionsj);

23: if all samples in chunk are processed then

24: w ← normalize_weights(w);

25: if Global_predictionss! = ys then

26: if num_experts(Experts) == l then

27: {Experts, w, expert_lifetime} ← Remove weakest expert ei based on experts score

28: end if

29: if num_experts(Experts) < l then

30: Expertsnew ← create_random_expert();

31: wnew ← 1/num_experts(Experts);

32: end if

33: end if

34: {Experts, w, expert_lifetime} ← Remove experts which score is below threshold θ

35: if num_experts(Experts) < k then

36: Expertsnew ← create_random_expert();

37: wnew ← 1/num_experts(Experts);

38: end if

39: end if

40: for i = 1,…, num_experts(Experts) do

41: Sample_weightss ← random_uniform_weight();

42: Expertsi ← learn_expert(Expertsi, xs, ys, Sample_weightss);

43: end for

44: return Global_predictions;

45: end for