Algorithm 1: BCA-Based Hyperparameter Optimization for CNN

Input:

D: Training dataset

M: CNN model structure

BCA_params: Algorithm parameters (bee_count, max_iterations, limit_T)

Output:

θ_best: Optimized hyperparameters (learning_rate, kernel_sizes, layer_depths)

W_best: Optimized weights

B_best: Optimized biases

1: Initialize bee population: scouts, employed bees, onlookers

2: Encode each bee as θ_i = (W_i, B_i)

3: Define fitness f(θ_i) = CrossEntropyLoss(M(θ_i), D)

4: for iteration = 1 to MaxIterations do

5://Employed bee phase

6: for each employed bee i do

7: Generate new candidate: θ_new,i = θ_i + φ_i · (θ_i − θ_k)

8: if f(θ_new,i) < f(θ_i) then θ_i ← θ_new,i

9: end for

10://Onlooker bee phase

11: Calculate selection probability P_i = f(θ_i)/Σ f(θ_j)

12: for each onlooker do

13: Select bee i with probability P_i

14: Perform local search as in Equation (5)

15: end for

16://Scout bee phase

17: for each solution unchanged for T iterations do

18: Replace with random new solution

19: end for

20://Elite preservation

21: Update global best if improved

22: end for

23: return W_best, B_best