tsb — Marketplace Fraud Triage

Skills you'll use: merge, derived columns, boolean filters, groupby().agg(), nlargestDataFrame, pivotTableFull.

1 · Join risk signals and score transactions

Blend transaction-level events with merchant risk metadata, then calculate expected loss and a risk score.

TypeScript

import { DataFrame, merge, nlargestDataFrame } from "tsb";

const transactions = DataFrame.fromRecords([
  { txn_id: "t001", merchant_id: "m01", segment: "electronics", device: "mobile", amount: 1299, card_testing: 0 },
  { txn_id: "t002", merchant_id: "m02", segment: "fashion", device: "desktop", amount: 184, card_testing: 0 },
  { txn_id: "t003", merchant_id: "m03", segment: "gaming", device: "mobile", amount: 720, card_testing: 1 },
  { txn_id: "t004", merchant_id: "m01", segment: "electronics", device: "desktop", amount: 899, card_testing: 0 },
  { txn_id: "t005", merchant_id: "m04", segment: "travel", device: "mobile", amount: 1450, card_testing: 1 },
  { txn_id: "t006", merchant_id: "m05", segment: "fashion", device: "mobile", amount: 92, card_testing: 0 },
  { txn_id: "t007", merchant_id: "m03", segment: "gaming", device: "desktop", amount: 610, card_testing: 0 },
  { txn_id: "t008", merchant_id: "m04", segment: "travel", device: "desktop", amount: 980, card_testing: 0 },
]);
const merchants = DataFrame.fromRecords([
  { merchant_id: "m01", merchant_name: "Northstar Cameras", chargeback_rate: 0.018 },
  { merchant_id: "m02", merchant_name: "Threadline", chargeback_rate: 0.012 },
  { merchant_id: "m03", merchant_name: "Pixel Arcade", chargeback_rate: 0.071 },
  { merchant_id: "m04", merchant_name: "FlyAway Deals", chargeback_rate: 0.096 },
  { merchant_id: "m05", merchant_name: "Studio Basics", chargeback_rate: 0.009 },
]);

const joined = merge(transactions, merchants, { on: "merchant_id", how: "left" });
const expectedLoss = joined.col("amount").mul(joined.col("chargeback_rate"));
const riskScore = expectedLoss.add(joined.col("card_testing").mul(25));
const triaged = joined.assign({ expected_loss: expectedLoss, risk_score: riskScore });

console.log(triaged.select(["txn_id", "merchant_name", "segment", "amount", "expected_loss", "risk_score"]).toString());

const bySegment = triaged
  .groupby("segment")
  .agg({ amount: "sum", expected_loss: "sum", risk_score: "mean" }, false)
  .sortValues("expected_loss", false);
console.log("\nSegment exposure:");
console.log(bySegment.toString());

const highRisk = triaged.filter(triaged.col("risk_score").ge(45));
console.log("\nManual review queue:");
console.log(nlargestDataFrame(highRisk, 3, { columns: "risk_score" }).select(["txn_id", "merchant_name", "risk_score"]).toString());

import pandas as pd

transactions = pd.DataFrame([
    {"txn_id": "t001", "merchant_id": "m01", "segment": "electronics", "device": "mobile", "amount": 1299, "card_testing": 0},
    {"txn_id": "t002", "merchant_id": "m02", "segment": "fashion", "device": "desktop", "amount": 184, "card_testing": 0},
    {"txn_id": "t003", "merchant_id": "m03", "segment": "gaming", "device": "mobile", "amount": 720, "card_testing": 1},
    {"txn_id": "t004", "merchant_id": "m01", "segment": "electronics", "device": "desktop", "amount": 899, "card_testing": 0},
    {"txn_id": "t005", "merchant_id": "m04", "segment": "travel", "device": "mobile", "amount": 1450, "card_testing": 1},
    {"txn_id": "t006", "merchant_id": "m05", "segment": "fashion", "device": "mobile", "amount": 92, "card_testing": 0},
    {"txn_id": "t007", "merchant_id": "m03", "segment": "gaming", "device": "desktop", "amount": 610, "card_testing": 0},
    {"txn_id": "t008", "merchant_id": "m04", "segment": "travel", "device": "desktop", "amount": 980, "card_testing": 0},
])
merchants = pd.DataFrame([
    {"merchant_id": "m01", "merchant_name": "Northstar Cameras", "chargeback_rate": 0.018},
    {"merchant_id": "m02", "merchant_name": "Threadline", "chargeback_rate": 0.012},
    {"merchant_id": "m03", "merchant_name": "Pixel Arcade", "chargeback_rate": 0.071},
    {"merchant_id": "m04", "merchant_name": "FlyAway Deals", "chargeback_rate": 0.096},
    {"merchant_id": "m05", "merchant_name": "Studio Basics", "chargeback_rate": 0.009},
])
triaged = transactions.merge(merchants, on="merchant_id", how="left")
triaged["expected_loss"] = triaged["amount"] * triaged["chargeback_rate"]
triaged["risk_score"] = triaged["expected_loss"] + triaged["card_testing"] * 25
print(triaged[["txn_id", "merchant_name", "segment", "amount", "expected_loss", "risk_score"]])
print(triaged.groupby("segment", as_index=False).agg({"amount":"sum", "expected_loss":"sum", "risk_score":"mean"}))
print(triaged[triaged["risk_score"] >= 45].nlargest(3, "risk_score"))

Click ▶ Run to execute

Ctrl+Enter to run · Tab to indent

2 · Build a fraud heatmap and merchant leaderboard

Summarize exposure by segment/device and rank merchants by estimated loss.

TypeScript

import { DataFrame, pivotTableFull } from "tsb";

const triaged = DataFrame.fromRecords([
  { merchant_id: "m01", segment: "electronics", device: "mobile", amount: 1299, expected_loss: 23.38 },
  { merchant_id: "m02", segment: "fashion", device: "desktop", amount: 184, expected_loss: 2.21 },
  { merchant_id: "m03", segment: "gaming", device: "mobile", amount: 720, expected_loss: 51.12 },
  { merchant_id: "m01", segment: "electronics", device: "desktop", amount: 899, expected_loss: 16.18 },
  { merchant_id: "m04", segment: "travel", device: "mobile", amount: 1450, expected_loss: 139.20 },
  { merchant_id: "m05", segment: "fashion", device: "mobile", amount: 92, expected_loss: 0.83 },
  { merchant_id: "m03", segment: "gaming", device: "desktop", amount: 610, expected_loss: 43.31 },
  { merchant_id: "m04", segment: "travel", device: "desktop", amount: 980, expected_loss: 94.08 },
]);

const heatmap = pivotTableFull(triaged, {
  index: "segment",
  columns: "device",
  values: "expected_loss",
  aggfunc: "sum",
  fill_value: 0,
  margins: true,
});
console.log("Expected loss by segment/device:");
console.log(heatmap.toString());

const merchantExposure = triaged
  .groupby("merchant_id")
  .agg({ amount: "sum", expected_loss: "sum" }, false)
  .sortValues("expected_loss", false);
console.log("\nMerchant exposure leaderboard:");
console.log(merchantExposure.toString());

import pandas as pd

triaged = pd.DataFrame([
    {"merchant_id": "m01", "segment": "electronics", "device": "mobile", "amount": 1299, "expected_loss": 23.38},
    {"merchant_id": "m02", "segment": "fashion", "device": "desktop", "amount": 184, "expected_loss": 2.21},
    {"merchant_id": "m03", "segment": "gaming", "device": "mobile", "amount": 720, "expected_loss": 51.12},
    {"merchant_id": "m01", "segment": "electronics", "device": "desktop", "amount": 899, "expected_loss": 16.18},
    {"merchant_id": "m04", "segment": "travel", "device": "mobile", "amount": 1450, "expected_loss": 139.20},
    {"merchant_id": "m05", "segment": "fashion", "device": "mobile", "amount": 92, "expected_loss": 0.83},
    {"merchant_id": "m03", "segment": "gaming", "device": "desktop", "amount": 610, "expected_loss": 43.31},
    {"merchant_id": "m04", "segment": "travel", "device": "desktop", "amount": 980, "expected_loss": 94.08},
])
heatmap = pd.pivot_table(
    triaged, index="segment", columns="device", values="expected_loss",
    aggfunc="sum", fill_value=0, margins=True,
)
print(heatmap)
print(triaged.groupby("merchant_id", as_index=False).agg({"amount":"sum", "expected_loss":"sum"}).sort_values("expected_loss", ascending=False))

Click ▶ Run to execute

Ctrl+Enter to run · Tab to indent

🛡️ Marketplace Fraud Triage

1 · Join risk signals and score transactions

2 · Build a fraud heatmap and merchant leaderboard