Predicting the Moment of Birth using Sensor Data in Dairy Cows
summary
Neck and leg accelerometer tags feed a Spark/Scala pipeline (Jupyter, Kubernetes, Azure) using event-driven schemas to predict calving from ~2,000 cows across eight farms and 600 million datapoints. A random-forest ensemble with leave-one-farm-out validation reaches ~80% overall accuracy and ~95% specificity but low sensitivity; imbalance and farm differences drive errors. Top predictor is lying-down percent change over a 7‑day window. Vendors supply black‑box activity labels; work continues on balancing, ensembles and neural nets.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:10Survey
- 6:05The truth about milk
- 16:15Sense of Sensor Project
- 17:53Sensor Data
- 22:02Timelapse
- 23:55Data Science Tools
- 26:30Typical Work
- 28:45Lag Activity
- 29:29Pipeline
- 32:55Future work
related talks
33mAnalytics in the age of the Internet of Things #bbuzz
38mRealtime prediction of C-section risk for laboring mothers
43mPredictive Maintenance: From POC to Production With Spark
37mForecasting using machine learning methods
41mBeefy ML: Ultra-Low Power Algorithms on Cattle
42mSo, You Want to "Do Big Data?"
41mBeefy ML: Ultra-Low Power Algorithms on Cattle
19mEnterprise AI-Powered Predictive Maintenance Platform