Reinforcement Learning in Business Applications
summary
Claims that reinforcement learning can tune fraud thresholds online using a 5‑dimensional hourly state (volume, fraud pressure, false positives, true positives, bank auth rate) and a binary action (raise/lower threshold), implemented with deep Q-learning in OpenAI Gym using TensorFlow + Keras-RL. Simulation uses price $100 with COGS 90 then 60; a frozen policy yields ~$3.49 → ~$22 per transaction, but online learning reoptimizes to about $30.25 while shifting false positives from ~40% to ~8% and fraud pressure up to ~25–28%, demonstrating practical gains.
AI-written summary · about
chapters (5)
related talks
13mHarnessing Machine Learning to Combat E-commerce Fraud
16mLeveraging ML to Create a Reliable Anti-Fraud System
26mAI-Powered Fraud Detection in Cloud-Native Digital Payments
14mCloud-Native AI for Fraud Prevention with Bayesian Networks
45mAnti-Fraud in Reverse and the Use of ML Methods in It
18mTransforming Financial Decision Intelligence Through Advanced NNs
37mThe Role of Machine Learning in Today’s Payment Ecosystems
29mDeep Learning Applications to Online Payment Fraud Detection