DecisionTreeClassifier (scikit-learn)

def fill_missing(df, ref):
    df = df.copy()
    df['payment_method'] = df['payment_method'].fillna('missing')
    for col in ['avg_watch_hours_30d', 'completion_rate_30d', 'buffering_ratio_30d', 'nps_score']:
        df[col] = df[col].fillna(ref[col].median())
    return df

train_proc = fill_missing(train_df, ref=train_df)
test_proc = fill_missing(test_df, ref=train_df)

drop_cols = ['customer_id', 'snapshot_date', 'churned_60d', 'payment_method']
cat_cols = ['plan_type', 'acquisition_channel', 'region', 'device_os',
            'contract_type', 'support_tier', 'usage_segment']
feature_cols = [c for c in train_proc.columns if c not in drop_cols]
num_cols = [c for c in feature_cols if c not in cat_cols]

preprocessor = ColumnTransformer([
    ('num', StandardScaler(), num_cols),
    ('cat', OneHotEncoder(handle_unknown='ignore'), cat_cols),
])
X_train_scaled = preprocessor.fit_transform(train_proc[feature_cols])
print('전처리 후 X_train shape:', X_train_scaled.shape)

전처리 후 X_train shape: (90000, 53)

from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier

X_train, X_val, y_train, y_val = train_test_split(X_train_scaled, y_train, test_size=0.2, random_state=42, stratify=y_train)

model_dt = DecisionTreeClassifier(max_depth=4, random_state=42, class_weight='balanced')
model_dt.fit(X_train, y_train)
pred = model_dt.predict(X_val)
print(confusion_matrix(y_val, pred))
print('F1 Score :', f1_score(y_val, pred))

[[10053  6020]
 [  575  1352]]
F1 Score : 0.29078395526400685

from sklearn.metrics import roc_auc_score

proba = model_dt.predict_proba(X_val)[:, 1]
roc_auc_score(y_val, proba)

0.707975