Algorithm 3 Federated Learning Algorithm

1: Input: Local datasets $D_i=(X_i,y_i)$ for clients i, $M_{missing}$, $O_{outliers}$, label encoder L   2: Output: Federated model ${\mathcal{M}}_{fed}$   3: procedure  Data Acquisition and Pre-processing   4:    for each client i do   5:        Load local dataset $D_i=(X_i,y_i)$   6:        Clean $D_i$ using $M_{missing}$, $O_{outliers}$   7:        Encode labels: ${\widehat{y}}_i=L\left(y_i\right)$   8:    end for   9:  end procedure 10: procedure  Local Model Training 11:    for each client i do 12:        Initialize local model: ${\mathcal{M}}_i$ 13:        Train ${\mathcal{M}}_i$ using DNN with $X_i$ and ${\widehat{y}}_i$ 14:    end for 15:  end procedure 16: procedure  Model Aggregation 17:    Initialize federated model: ${\mathcal{M}}_{fed}$ 18:    Aggregate local models: ${\mathcal{M}}_{fed}=\mathrm{Aggregate}\_\mathrm{Models}\left(\{{\mathcal{M}}_1,{\mathcal{M}}_2,\dots ,{\mathcal{M}}_n\}\right)$ 19:  end procedure 20: procedure  Global Model Training 21:    Compile ${\mathcal{M}}_{fed}$ with optimizer (e.g., Adam) and loss (e.g., Sparse Categorical Cross-Entropy) 22:    Train ${\mathcal{M}}_{fed}$ using global data 23:  end procedure 24: procedure  Final Classifier 25:    Output: Final classifier ${\mathcal{M}}_{fed}$ 26:  end procedure 27:  Return ${\mathcal{M}}_{fed}$