Algorithm 1 Genetic-Embedded Nuclear Reaction Optimization (GNR) Algorithm

Require: Dataset D, Population size N = 500, Max generations T = 30, Early stopping patience P = 5 Ensure: Optimized subset of gene                        ▷ Preprocessing and Filtering 1:  Handle missing values (mean imputation) 2:  Normalize features using Z-score 3:  Encode categorical labels (if any) 4:  Compute F-score for each gene 5:  Select top 500 genes based on F-score                               ▷ Initialization 6:   for each subset size k ∈ {5, 10, 15} do 7:      Initialize population of continuous vectors xi ∈ [0, 1]D 8:      Randomly initialize each xi in population 9:      Evaluate fitness using SVM with LOOCV 10:    Set $X_{\mathit{best}}$ ← best solution, no_improve ← 0                  ▷ Fission Phase: Exploration via perturbation 11:    for g = 1 to T do 12:    for each solution $x_i$ do 13:       Generate new solution ${x_i}^{Fi}$ using Equation (2) 14:       Compute dynamic step sizes using Equations (3) and (4) 15:       Apply mutation factors using Equations (5) and (6) 16:    end for             ▷ Fusion Phase: Exploitation with embedded crossover 17:    for each solution $x_i$ do 18:       Apply ionization via Equation (7) 19:       if solutions are similar then 20:       Apply Lévy flight adjustment using Equation (8) 21:       end if 22:       Generate ${x_i}^{Fu}$ as follows: 23:       if rand < 0.3 then 24:       Select random solution $x_{rand}$ 25:       for each gene d ∈ {1, . . . , D} do 26:          ${x_i}^{Fu} \left[d\right]$ ← if rand < 0.8: $X_{\mathit{best}} \left[d\right]$; else: $x_{rand} \left[d\right]$ 27:       end for 28:       else 29:       Apply fusion using Equation (9) or (10) depending on similarity 30:       end if 31:    end for                 ▷ Fitness Evaluation and Best Solution Update 32:    for each solution $x_i$ do 33:       Evaluate fitness via SVM + LOOCV 34:       if $Fitness\left(x_i\right)$ > $Fitness\left(X_{\mathit{best}}\right)$ then 35:       Update $X_{\mathit{best}}$ ← $x_i$, reset no_improve ← 0 36:       else 37:       Increment no_improve ← no_improve + 1 38:       end if 39:    end for 40:    if no_improve ≥ $P$ then 41:       break 42:    end if 43:    end for 44:   end for                                ▷ Final Output 45:   return best gene subset $X_{\mathit{best}}$