Using Changepoint and Bayesian Analysis to Drive Safety Improvements in Mining
summary
Claims and results are concrete: changepoint detection with PELT with Poisson changepoint finds few shifts, Bayesian MCMC gives confidence about those shifts, and the operational output becomes a monthly probability-of-change alerts system moving toward weekly alerts. Tools used include a Poisson model, PELT, and MCMC via arbist. The surprising result is stability: incidents sit around one per month and the Bayesian posterior shows ~98% probability of a flat slope, so the pipeline yields actionable uncertainty rather than fuzzy anecdotes.
AI-written summary · about
chapters (8)
- 0:00Safety in mining operations
- 3:56Measuring safety improvements
- 7:25Traditional changepoint limits
- 11:38Bayesian changepoint analysis
- 13:22Beyond mining applications
- 14:49Key takeaways
- 18:22Complexity for better decisions
- 19:10Q&A
related talks
17mStopping Performance Regression via Changepoint Detection
28mOnline Change Point Detection Using Spark Streaming
50mQuantifying Swiss Cheese, the Bayesian Way
47mDifferentially Private Change-Point Detection
21mHazards on the Causal Path
49mShifting Safety Techniques To a Statistical World
01h 10mBayesian Optimization and Machine Learning for Accelerating Experiments in the Physical Sciences
01h 27mProbabilistic Python: An Introduction to Bayesian Modeling with PyMC