Train score → Excellent
Validation  → Poor

Likely overfitting।

Possible solutions:

Regularization
Simpler model
More data
Feature selection
Reduce tree depth
Early stopping
Dropout
Cross-validation