train_ml_model.py

import pickle
import os
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split

np.random.seed(42)

# Нормальные транзакции
normal_amounts = np.random.uniform(100, 10000, 60)
normal_hours = np.random.randint(8, 22, 60)  # Днем
normal_weekend = np.random.choice([0, 1], 60, p=[0.7, 0.3])
normal_age = np.random.uniform(180, 3650, 60)  # 6 мес - 10 лет

# Мошеннические транзакции
fraud_amounts = np.random.uniform(50000, 500000, 40)
fraud_hours = np.random.choice([0, 1, 2, 3, 22, 23], 40)  # Ночью
fraud_weekend = np.random.choice([0, 1], 40, p=[0.3, 0.7])
fraud_age = np.random.uniform(1, 30, 40)  # Очень новые аккаунты

# Объединяем
X_normal = np.column_stack([normal_amounts, normal_hours, normal_weekend, normal_age])
X_fraud = np.column_stack([fraud_amounts, fraud_hours, fraud_weekend, fraud_age])

X = np.vstack([X_normal, X_fraud])
y = np.hstack([np.zeros(60), np.ones(40)])

# Разделяем на train/test
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

# Нормализуем
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)

#  Используем Random Forest
model = RandomForestClassifier(
    n_estimators=50, max_depth=5, random_state=42  # 50 деревьев
)
model.fit(X_train_scaled, y_train)

# Оцениваем точность
train_acc = model.score(X_train_scaled, y_train)
test_acc = model.score(X_test_scaled, y_test)

# Сохраняем
model_data = {
    "model": model,
    "scaler": scaler,
    "version": "1.0",
    "train_accuracy": train_acc,
    "test_accuracy": test_acc,
    "features": ["amount", "hour_of_day", "is_weekend", "account_age_days"],
    "trained_at": "2025-10-24",
}


os.makedirs("app/static", exist_ok=True)

with open("app/static/fraud_model_v1.pkl", "wb") as f:
    pickle.dump(model_data, f)

print("✅ Model trained and saved to app/static/fraud_model_v1.pkl")
print(f"📊 Train accuracy: {train_acc:.2%}")
print(f"📊 Test accuracy: {test_acc:.2%}")
print(f"🌲 Model: RandomForest with {getattr(model, 'n_estimators', 'unknown')} trees")
print(f"📋 Features: {model_data['features']}")

test_example = scaler.transform([[150000, 3, 1, 2]])
fraud_prob = model.predict_proba(test_example)[0][1]
print("\n🔍 Test prediction:")
print("   Transaction: 150,000 руб at 3 AM, weekend, 2 days old account")
print(f"   Fraud probability: {fraud_prob:.2%}")