AI Powered Root Cause Analysis for Production Alerts
summary
PayPal trains a multiclass classifier on consolidated alert and incident-ticket data from ~2,500 apps, using Jupyter and scikit-learn with features like alert message, service, region, timestamp and closure comments, and selects a gradient boosting classifier that reaches about 82% accuracy on test data. Models deploy as a REST API via Python Flask, predictions are compared to labeled root cause fields and retrained periodically to improve noisy labels and data hygiene.
AI-written summary · about
chapters (4)
- 0:00Introduction to RCA
- 5:22Why machine learning
- 15:01Model training and deployment
- 17:07Real-time evaluation and improvements
related talks
12mUnleashing the Power of AI for Performance Monitoring& Anomaly Detection
25mSymptom-based Alerting for Machine Learning - What I Learned...
30mMassive Scale Anomaly Detection Framework
08mAI-Powered Fraud Detection in Financial System
22mAutomated Troubleshooting of Live Site Issues
01h 30mAI in Finance From Insight to Action
36mContinuous Monitoring of Machine Learning Models
19mThe Rise of AI in Anomaly Detection