支付平台用户流失预测（XGBoost）

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split

# Load dataset
df = pd.read_csv('allinpay_user_data.csv', parse_dates=['registration_date'])
print(f"Dataset shape: {df.shape}")  # (50000, 15)
print(f"Churn rate: {df['is_churned'].mean():.2%}")  # 12.00%

# Train/test split with stratification (preserve class ratio)
X = df.drop(['user_id', 'is_churned'], axis=1)
y = df['is_churned']
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42, stratify=y
)
print(f"Train: {X_train.shape[0]}, Test: {X_test.shape[0]}")  # 40000, 10000

# RFM Feature Engineering
df['recency'] = (pd.Timestamp.now() - df['last_transaction_date']).dt.days
df['frequency'] = df.groupby('user_id')['transaction_id'].transform('count')
df['monetary'] = df.groupby('user_id')['transaction_amount'].transform('sum')

# Quintile scoring (1=worst, 5=best)
df['r_score'] = pd.qcut(df['recency'], 5, labels=[5,4,3,2,1]).astype(int)
df['f_score'] = pd.qcut(df['frequency'].rank(method='first'), 5, labels=[1,2,3,4,5]).astype(int)
df['m_score'] = pd.qcut(df['monetary'].rank(method='first'), 5, labels=[1,2,3,4,5]).astype(int)

# Behavioral: Activity trend detection
df['session_trend'] = df['app_sessions_7d'] / (df['app_sessions_30d'] / 4.28 + 1e-6)
df['activity_decay_rate'] = np.polyfit(range(8), weekly_activity_series, 1)[0]

# Engagement delta features
df['login_frequency_change'] = (
    df['login_count_recent_14d'] / (df['login_count_prior_14d'] + 1e-6) - 1
)
df['transaction_gap_increase'] = (
    df['avg_gap_recent_30d'] - df['avg_gap_prior_30d']
)

import xgboost as xgb
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import roc_auc_score, classification_report

# Handle class imbalance with scale_pos_weight
scale_ratio = (y_train == 0).sum() / (y_train == 1).sum()  # ~7.33

# XGBoost with hyperparameter tuning
param_grid = {
    'max_depth': [4, 6, 8],
    'learning_rate': [0.01, 0.05, 0.1],
    'n_estimators': [200, 500, 800],
    'min_child_weight': [1, 3, 5],
    'subsample': [0.8, 0.9],
    'colsample_bytree': [0.8, 0.9],
}

xgb_clf = xgb.XGBClassifier(
    objective='binary:logistic',
    scale_pos_weight=scale_ratio,
    eval_metric='auc',
    random_state=42,
    use_label_encoder=False,
)

grid_search = GridSearchCV(
    xgb_clf, param_grid, scoring='roc_auc',
    cv=5, n_jobs=-1, verbose=1
)
grid_search.fit(X_train, y_train)

# Best model evaluation
best_model = grid_search.best_estimator_
y_pred_proba = best_model.predict_proba(X_test)[:, 1]
print(f"AUC-ROC: {roc_auc_score(y_test, y_pred_proba):.4f}")  # 0.8903