Finding Driving Style Patterns in Caterpillar Machine Data
summary
A working pipeline extracts engineered powertrain features from 10‑minute windows of Caterpillar telematics, reduces 50–60 features to ~8 principal components using PCA to 80% variance, and runs K‑means across K=5–20 (team selects K=13) to produce a few dominant driving patterns; implementation uses HDF5 storage and a HDF5 + Celery parallel pipeline in Python, processes ~150,000 windows (runtime drops from ~48h single‑node to ~4h on 12 cores), and yields believable clusters for control‑software validation.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 5:50Manual pattern selection problem
- 9:44Automated pattern discovery
- 13:17Feature engineering & data reduction
- 19:38Data storage & scalability
- 24:55Parallel computing & workflow optimization
- 34:28Results validation & insights
- 40:13Project timeline & applications
- 43:07Future directions & Q&A
related talks
41mClassifying train and car journeys using telematics data
38mTransforming Data to Unlock Its Latent Value
40mSmart cars of tomorrow: real-time driving patterns
45mForecasting at Scale
32mTesting Machine Learning Systems: What it is and why you should care
23mDataEngBytes 2023 - MEL-T3-01 - Rajath Vanikul
57mGood Practices for Applied Machine Learning
24mThe TL;DR of EDA