import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.feature_selection import mutual_info_classif

np.random.seed(42)

n = 5000
df = pd.DataFrame({
    "avg_transaction_value": np.random.normal(150, 40, n),
    "num_transactions_30d": np.random.poisson(12, n),
    "account_age_days": np.random.exponential(400, n),
    "device_risk_score": np.random.beta(2, 5, n),
})

df["target"] = (df["device_risk_score"] * 3 + np.random.normal(0, 0.3, n) > 1.2).astype(int)
# this is the kind of feature that sneaks into real datasets constantly -
# a flag count that got backfilled after the outcome was already known
df["post_hoc_flag_count"] = df["target"] * np.random.poisson(2, n)

X = df.drop(columns=["target"])
y = df["target"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)