File size: 6,644 Bytes
6e27e66 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 | import numpy as np
import pandas as pd
import random
import string
import json
import joblib
from pathlib import Path
# ─── Load artifacts ───────────────────────────────────────────────────────────
BASE = Path(__file__).parent
with open(BASE / "config.json") as f:
CONFIG = json.load(f)
W_STAR = CONFIG["optimal_weight"] # 0.3
THR_BLOCK = CONFIG["threshold_layer3"]["block"] # 0.8
THR_ALERT = CONFIG["threshold_layer3"]["alert_otp"] # 0.5
FEAT_A = CONFIG["feature_list"]["model_a"]
FEAT_B = CONFIG["feature_list"]["model_b"]
MODEL_A = joblib.load(BASE / "model_a_pipeline.joblib")
MODEL_B = joblib.load(BASE / "model_b_pipeline.joblib")
PAYEE_PROFILE = pd.read_csv(BASE / "payee_profile_store.csv")
# Normalise column name — support both 'mule_account_id' and 'nameDest'
_id_col = "mule_account_id" if "mule_account_id" in PAYEE_PROFILE.columns else PAYEE_PROFILE.columns[0]
PAYEE_PROFILE = PAYEE_PROFILE.set_index(_id_col)
# ─── Helpers ──────────────────────────────────────────────────────────────────
def _rand_id(prefix: str, n: int = 4) -> str:
return prefix + "".join(random.choices(string.digits, k=n))
def generate_accounts(n: int = 5, seed: int = None) -> list[dict]:
"""
Create n random sender accounts with a risk profile each.
Called once per simulation session.
"""
if seed is not None:
random.seed(seed)
np.random.seed(seed)
profiles = []
for _ in range(n):
risk = random.choices(
["low", "medium", "high"], weights=[0.60, 0.28, 0.12]
)[0]
profiles.append(
{
"account_id": _rand_id("ACC-"),
"risk_level": risk,
"balance": round(random.uniform(5_000, 500_000), 2),
"avg_tx_amount": round(random.uniform(1_000, 80_000), 2),
"tx_history_steps": random.randint(50, 400),
"accessibility_prob": {"low": 0.04, "medium": 0.15, "high": 0.55}[risk],
"remote_app_prob": {"low": 0.02, "medium": 0.08, "high": 0.40}[risk],
"status": "normal",
}
)
return profiles
def _lookup_payee(dest_id: str) -> dict:
"""Return payee profile row or cold-start defaults."""
if dest_id in PAYEE_PROFILE.index:
row = PAYEE_PROFILE.loc[dest_id]
return {
"dest_cashout_ratio": float(row.get("dest_cashout_ratio", 0.0)),
"dest_avg_step_to_cashout": float(row.get("dest_avg_step_to_cashout", 0.0)),
"dest_tx_count_received": float(row.get("dest_tx_count_received", 0.0)),
"dest_is_new_account": int(row.get("dest_is_new_account", 1)),
}
return {
"dest_cashout_ratio": 0.0,
"dest_avg_step_to_cashout": 0.0,
"dest_tx_count_received": 0.0,
"dest_is_new_account": 1,
}
def generate_transaction(account: dict, step: int, tx_counter: int) -> dict:
"""
Simulate one TRANSFER transaction for the given account.
Returns a feature dict ready for model scoring.
"""
risk = account["risk_level"]
avg = account["avg_tx_amount"]
# Device signals
is_acc = int(random.random() < account["accessibility_prob"])
is_rem = int(random.random() < account["remote_app_prob"])
hesit = round(np.clip(np.random.lognormal(3.2 if risk == "low" else 4.0, 0.9), 5, 300), 1)
# Velocity features
tx_count_per_step = max(1, int(np.random.poisson(2 if risk == "low" else 4)))
dest_tx_count_24step = max(0, int(np.random.poisson(1 if risk == "low" else 3)))
# Amount anomaly
multiplier = np.random.lognormal(0, 0.4) if risk == "low" else np.random.lognormal(0.8, 0.9)
amount = round(avg * multiplier, 2)
amount_vs_avg_ratio = round(amount / avg, 4)
# Temporal
hour = random.randint(0, 23)
is_odd_hours = int(hour < 6 or hour >= 22)
# Recipient
is_first_time = int(random.random() < (0.10 if risk == "low" else 0.45))
# Pick a payee — sometimes from profile store, sometimes new
if random.random() < 0.3 and len(PAYEE_PROFILE) > 0:
dest_id = random.choice(PAYEE_PROFILE.index.tolist())
else:
dest_id = _rand_id("C", 6)
payee = _lookup_payee(dest_id)
return {
"tx_id": f"TX-{tx_counter:05d}",
"step": step,
"account_id": account["account_id"],
"dest_id": dest_id,
"amount": amount,
"hour": hour,
# Model A features
"is_accessibility_enabled": is_acc,
"is_remote_app_running": is_rem,
"hesitation_time_sec": hesit,
"tx_count_per_step": tx_count_per_step,
"dest_tx_count_24step": dest_tx_count_24step,
"amount_vs_avg_ratio": amount_vs_avg_ratio,
"is_odd_hours": is_odd_hours,
"is_first_time_recipient": is_first_time,
# Model B features (payee profile)
**payee,
}
def score_transaction(tx: dict) -> dict:
"""
Run 3-layer scoring pipeline on a single transaction dict.
Returns tx enriched with scores and decision.
"""
row_a = pd.DataFrame([{f: tx[f] for f in FEAT_A}])
row_b = pd.DataFrame([{f: tx[f] for f in FEAT_B}])
prob_a = float(MODEL_A.predict_proba(row_a)[0, 1])
prob_b = float(MODEL_B.predict_proba(row_b)[0, 1])
# Layer 1 — Hard Rules
rule1 = tx["is_accessibility_enabled"] == 1 and tx["is_remote_app_running"] == 1
rule2 = tx["hesitation_time_sec"] > 45 and tx["is_first_time_recipient"] == 1
# Ensemble score
final_score = W_STAR * prob_a + (1 - W_STAR) * prob_b
# Rule 2 bias
if rule2:
final_score = max(final_score, 0.6)
# Layer 3 — Threshold decision
if rule1:
decision = "Block (Rule 1)"
layer = "Layer 1"
elif final_score >= THR_BLOCK:
decision = "Block"
layer = "Layer 3"
elif final_score >= THR_ALERT:
decision = "Alert + OTP"
layer = "Layer 3"
else:
decision = "Pass"
layer = "Layer 3"
return {
**tx,
"prob_a": round(prob_a, 4),
"prob_b": round(prob_b, 4),
"final_score": round(final_score, 4),
"rule1": rule1,
"rule2": rule2,
"decision": decision,
"layer": layer,
}
|