Algorithm A1: Machine learning pipeline for model-type and aging-type classification

Input: LSR dataset $D_{\mathrm{LSR}}=\{X_{\mathrm{LSR}},y_{\mathrm{model}\_\mathrm{LSR}},y_{\mathrm{aging}\_\mathrm{LSR}}\}$, EPDM dataset $D_{\mathrm{EPDM}}=\{X_{\mathrm{EPDM}},y_{\mathrm{model}\_\mathrm{EPDM}},y_{\mathrm{aging}\_\mathrm{EPDM}}\}$    Output: Final accuracy $A_{\mathrm{final}}$, classification report, and confusion matrix $C_{\mathrm{final}}$ 1 Load datasets $D_{\mathrm{LSR}}$ and $D_{\mathrm{EPDM}}$ from CSV files; 2 Concatenate features: $X_{\mathrm{model}}\leftarrow [X_{\mathrm{LSR}},X_{\mathrm{EPDM}}]$; 3 Concatenate model-type labels: $y_{\mathrm{model}}\leftarrow [y_{\mathrm{model}\_\mathrm{LSR}},y_{\mathrm{model}\_\mathrm{EPDM}}]$; 4 Encode model-type labels: $y_{\mathrm{model}\_\mathrm{encoded}}\leftarrow \mathrm{LE}\left(y_{\mathrm{model}}\right)$; 5 Normalize features: $X_{\mathrm{model}\_\mathrm{scaled}}\leftarrow \mathrm{SC}\left(X_{\mathrm{model}}\right)$; 6 Concatenate features: $X_{\mathrm{aging}}\leftarrow [X_{\mathrm{LSR}},X_{\mathrm{EPDM}}]$; 7 Concatenate aging-type labels: $y_{\mathrm{aging}}\leftarrow [y_{\mathrm{aging}\_\mathrm{LSR}},y_{\mathrm{aging}\_\mathrm{EPDM}}]$; 8 Encode aging-type labels: $y_{\mathrm{aging}\_\mathrm{encoded}}\leftarrow \mathrm{LE}\left(y_{\mathrm{aging}}\right)$; 9 Normalize features: $X_{\mathrm{aging}\_\mathrm{scaled}}\leftarrow \mathrm{SC}\left(X_{\mathrm{aging}}\right)$; 10 Split $X_{\mathrm{model}\_\mathrm{scaled}},y_{\mathrm{model}\_\mathrm{encoded}}$ and $X_{\mathrm{aging}\_\mathrm{scaled}},y_{\mathrm{aging}\_\mathrm{encoded}}$ into 80–20 training and testing sets;     Input: Model M, training and testing sets for model-type or aging-type classification task     Output: Performance metrics: accuracy A, classification report, and confusion matrix C 11 Train model M on training data $X_{\mathrm{train}},y_{\mathrm{train}}$ to minimize the loss function L; 12 $$M\leftarrow argminL(M\left(X_{\mathrm{train}}\right),y_{\mathrm{train}})$$    Predict test labels ${\widehat{y}}_{\mathrm{test}}$ on $X_{\mathrm{test}}$:; 13 $${\widehat{y}}_{\mathrm{test}}=M\left(X_{\mathrm{test}}\right)$$    Calculate accuracy A:; 14 $$A={\displaystyle \frac{1}{n}}\sum _{i=1}^n⊮({\widehat{y}}_i=y_i)$$     Output classification report and confusion matrix C; 15 Define and initialize models for both classification tasks as follows: XGBoost: Define with default hyperparameters. Multilayer Perceptron (MLP): Define with 100 hidden units and ReLU activation. Deep Neural Network (DNN): Define with hidden layers $[150,100,50]$. Stacked Model (RF + SVM): Use Random Forest (RF) and Support Vector Machine (SVM) as base estimators, with an RF as the final estimator.     Input: Stacked model $M_{\mathrm{stacked}}$, Parameter grid G     Output: Optimal parameters $G^{*}$ and best model $M_{\mathrm{best}}$ 16 Perform a randomized search on $M_{\mathrm{stacked}}$ with grid G; 17 for each parameter configuration $g\in G$ do$\phantom{(}$ 18 ${\displaystyle ˪}$ Evaluate $M_{\mathrm{stacked}}\left(g\right)$ on the training set and compute accuracy $A\left(g\right)$; 19 Set $G^{*}\leftarrow argmaxA\left(g\right)$ and update best model $M_{\mathrm{best}}=M_{\mathrm{stacked}}\left(G^{*}\right)$;    Input: Optimized stacked model $M_{\mathrm{best}}$, testing set $X_{\mathrm{test}},y_{\mathrm{test}}$    Output: Final accuracy $A_{\mathrm{final}}$, classification report, and confusion matrix $C_{\mathrm{final}}$ 20 Use $M_{\mathrm{best}}$ to predict labels ${\widehat{y}}_{\mathrm{test}}$ on test data:; 21 $${\widehat{y}}_{\mathrm{test}}=M_{\mathrm{best}}\left(X_{\mathrm{test}}\right)$$    Calculate and output final accuracy $A_{\mathrm{final}}$, classification report, and confusion matrix $C_{\mathrm{final}}$; 22 return $A_{\mathrm{final}},C_{\mathrm{final}}$