bandicoot: a toolbox to analyze mobile phone metadata
summary
Bandicoot is a free open-source Python toolbox that computes 160+ behavioral indicators from cellphone CDRs or CSV logs, plus reporting variables and weekly/monthly aggregations, and exports JSON or CSV for ML. It runs without heavy dependencies (MIT license), offers an Android app to dump personal logs as CSV, and supports carrier-side privacy-preserving feature export. Papers using it report predictions like ~63% accuracy on some personality dimensions and ~75–88% gender prediction depending on sample size.
AI-written summary · about
chapters (23)
- 0:00Mobile phone metadata analysis
- 1:53Accessing metadata challenges
- 3:39Introducing Bandicoot
- 5:06Bandicoot metadata processing
- 8:20Data collection with apps
- 10:12User interaction patterns
- 11:14Computing features with Bandicoot
- 13:04Feature granularity and stats
- 16:01Custom indicators in Bandicoot
- 17:03Reproducibility and privacy
- 18:49Personality traits prediction
- 21:59Gender distribution prediction
- 23:05Bandicoot open-source release
- 24:34Partnerships with Orange
- 25:57Android vs iOS data access
- 27:26Mobile providers as data sources
- 28:12Bandicoot design and docs
- 29:21Facebook and chat data
- 30:29Python version compatibility
- 31:34Large dataset performance
- 32:45Future sample datasets
- 33:06Homeless mobility case study
- 34:09Real-time movement tracking
related talks
28mYour data, your insights: creating personal data projects to (re-)own the...
45mMining dockless bikeshare and dockless scootershare trip data
28mMall Analytics Using Telco Data & Pandas
24mSirbarksalot Bark Detection in Python
41mThe What, Why and How of Web Analytics Testing
02mDBTB INT Aaron Kalb r
32mBuild a data visualization app for your phone
20mBI Toolkit Powered by Natural Language Processing