classifier/tune_xgboost.py

import numpy as np
import xgboost as xgb
from sklearn.metrics import accuracy_score, classification_report
from sklearn.model_selection import ParameterGrid
import joblib

X_train = np.load("featureextraction/final_features/train_X.npy")
y_train = np.load("featureextraction/final_features/train_y.npy")

X_val = np.load("featureextraction/final_features/val_X.npy")
y_val = np.load("featureextraction/final_features/val_y.npy")

print("Train shape:", X_train.shape)
print("Val shape:", X_val.shape)

param_grid = {
    "n_estimators": [300, 500],
    "max_depth": [4, 6, 8],
    "learning_rate": [0.03, 0.05, 0.1],
    "subsample": [0.8, 1.0],
    "colsample_bytree": [0.8, 1.0]
}

best_acc = 0.0
best_model = None
best_params = None

print("\n Starting XGBoost Hyperparameter Tuning...\n")

for idx, params in enumerate(ParameterGrid(param_grid), start=1):
    print(f"Trial {idx}: {params}")

    model = xgb.XGBClassifier(
        objective="binary:logistic",
        eval_metric="logloss",
        random_state=42,
        **params
    )

    model.fit(X_train, y_train)

    val_preds = model.predict(X_val)
    acc = accuracy_score(y_val, val_preds)

    print("Validation Accuracy:", round(acc, 4))

    if acc > best_acc:
        best_acc = acc
        best_model = model
        best_params = params
        print("New best model found")

    print("-" * 60)

print("\n🏆 BEST VALIDATION ACCURACY:", round(best_acc, 4))
print("🏆 BEST PARAMETERS:", best_params)

final_preds = best_model.predict(X_val)

print("\nFINAL VALIDATION REPORT\n")
print(
    classification_report(
        y_val,
        final_preds,
        target_names=["Human", "AI"]
    )
)

best_model.save_model("classifier/xgboost_final_model.json")
joblib.dump(best_model, "classifier/xgboost_final_model.pkl")

print("\n Tuned XGBoost model saved successfully")