santander-customer-prediction

Santander Customer Transaction Prediction

Modelo para predição de transações de clientes baseado no dataset do Kaggle Santander Customer Transaction Prediction.

Este modelo foi treinado usando Gradient Boosting Classifier com redução de dimensionalidade via PCA.

Metrics

• Accuracy: 0.9138 (91.38%)
• Roc Auc: 0.6275 (62.75%)
• Precision: 0.6777
• Recall: 0.2693
• F1 Score: 0.3855

Deployment Criteria

This model was automatically deployed because it meets the following criteria:

• Accuracy > 90%
• ROC AUC Score > 60%

Only models that meet these performance thresholds are deployed to ensure quality.

Usage

from huggingface_hub import hf_hub_download
import joblib

# Download
model_path = hf_hub_download(repo_id="your-repo", filename="model.pkl")
pca_path = hf_hub_download(repo_id="your-repo", filename="pca.pkl")
scaler_path = hf_hub_download(repo_id="your-repo", filename="scaler.pkl")

# Load
model = joblib.load(model_path)
pca = joblib.load(pca_path)
scaler = joblib.load(scaler_path)

# Predict (assume X is your input data)
X_scaled = scaler.transform(X)
X_transformed = pca.transform(X_scaled)
predictions = model.predict(X_transformed)
probabilities = model.predict_proba(X_transformed)

Model Architecture

• Algorithm: Gradient Boosting Classifier
• Dimensionality Reduction: PCA (95% variance retained)
• Preprocessing: Standard Scaler
• Features: 200 anonymous features from Santander dataset

Training Details

• Train/Test Split: 80/20
• Random State: 42 (for reproducibility)
• Cross-validation: Used during hyperparameter tuning