Urgent.News

What's breaking now, across thousands of outlets.

Tech

Why 0.50 Thresholds Burn Money in Banking (And How We Built a Prescriptive ML Engine)

Most machine learning tutorials treat classification like a neat academic contest. You download a dataset, split it eighty-twenty, run a gradient booster, look at an ROC-AUC curve, and call model.predict() . If you do that inside a retail bank, you quietly burn money. Here is why. Standard predict() calls in Scikit-learn and XGBoost apply a hardcoded zero point five probability cutoff. That…

In the banking industry, classification algorithms are often used to predict customer churn. However, machine learning tutorials typically use a 0.50 probability cutoff, assuming that a false positive and false negative cost the same. This assumption is not true in banking, where false positives can be much more costly. Banks lose money when customers leave, eroding net interest margin, card swipe fees, and future underwriting opportunities.

The expected profit from retention efforts is calculated as (p × r × V) - C, where p is the churn probability, r is the acceptance rate, and V is the customer lifetime value. According to the given numbers, the profit-maximizing boundary for churn probability is 33.33%, not 50%. Traditional machine learning approaches using a 0.50 cutoff miss over 45% of genuine churners and waste budget on uncalibrated guesses.

The prescriptive machine learning engine builds a decision boundary based on the bank's unit economics, preserving genuine churners while keeping watchlists for those who may leave. This approach ensures that outreach efforts are only made when they are expected to be profitable, maximizing the return on investment.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

Check Point Patches a Root Daemon That Accepts Commands From Any Local Process

TL;DR CVE-2026-123039 is a local privilege escalation in Gaia OS, the operating system that runs on Check Point security appliances, where a daemon named dynamic_server executes commands as root.

  • Local privilege escalation vulnerability in Gaia OS, CVE-2026-123039
  • Dynamicserver daemon executes commands as root via world-writable Unix socket
  • Check Point released hotfixes; exploit active in several dozen customer regions

More from Tuesday 17 November →