Machine Learning Model of Paper Titled "Tracing Cretaceous Ridge Subduction Beneath Eastern Asian Continent Using Machine Learning"

Research Hypothesis
First, we need to train the machine‑learning models from documented geochemical data of ridge-subduction-related igneous rocks. Then we applied the model to test igneous rocks in tagert areas, and the geochemical fingerprints of ridge subduction can be detected through machine learning models trained on major‑ and trace‑element data.

Data Description
The dataset of documented ridge-subduction-related igneous rocks and igneous rocks from the Eastren Asian continental margin are from Supporting Information.
Training dataset: A binary label (0 = no ridge‑subduction signal; 1 = ridge‑subduction signal) assigned based on independent geological evidence.

How to Run the Model
Update the file paths in the Python code to point to these Excel files on your computer. Execute the code. The script will train Random‑Forest and Gradient‑Boosting classifiers, evaluate them on a tagert test set, and generate five output figures plus a prediction file.

Key Outputs
Feature‑importance plots: Show which geochemical parameters are most diagnostic of ridge subduction.
Correlation heatmap: Reveals interdependencies among the 13 features.
ROC‑curve comparison: Illustrates the trade‑off between true‑positive and false‑positive rates for both models.
Confusion matrices: Detail the number of correct/incorrect predictions for each class.
Prediction statistics: Provide the predicted labels and probabilities for new samples.

Interpretation & Use
The models serve as a quantitative tool for identifying ridge‑subduction signatures in Cretaceous igneous rocks of East Asia. High scores on the test set indicate that the selected geochemical features contain systematic patterns related to ridge subduction. The output figures help visualize model performance and the relative importance of each geochemical indicator. Users can apply the trained models to new, unlabeled geochemical data from the same region and age range to obtain probabilistic predictions of ridge‑subduction affinity.