Context

ShopShield runs a binary classification model that flags suspicious e-commerce orders for manual review before fulfillment. The customer success team says enterprise merchants keep focusing on the model's 96.2% accuracy, while complaining that too many legitimate orders are being flagged and delayed.

Current Performance

Confusion Matrix Snapshot

The Problem

A large merchant asks: "If your model is over 96% accurate, why are 2,375 good orders being flagged?" You need to explain this clearly, quantify the tradeoff, and recommend whether the current threshold is appropriate.

Requirements

1. Explain why high accuracy can coexist with a large number of false positives.
2. Interpret the confusion matrix in customer-friendly business terms.
3. Compare the current threshold with the previous threshold and describe the tradeoff.
4. Recommend how you would communicate the right primary metric for this use case.
5. Suggest concrete steps to reduce false positives without sharply increasing missed fraud.

Constraints

• Fraud prevalence is low: 800 fraudulent orders out of 50,000 daily orders.
• Each false positive delays shipment and costs the merchant about $4 in support and handling.
• Each false negative costs about $120 in fraud loss.
• Manual review capacity is capped at 3,200 orders per day.