Using EEG and Azure Machine Learning to Perform Lie Detection
summary
Uses an Emotiv EPOC 14-channel headset at roughly 128 Hz sampling to record labeled truth/lie trials (eyes closed, markers for start/end) converted to CSV, feature‑extracted (wavelets, ERP/P300) and trained in Azure Machine Learning with a 70/30 train/test split. An early decision‑jungle model reports ~93% accuracy and AUC ~98% on personalized datasets (husband, manager), but dataset is small, risks overfitting, and author notes need for neutral baselines and substantially more subjects before claiming general lie‑detection performance.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:20Electrochemical Basics
- 4:22What got me excited about this headset
- 5:10The Control Panel
- 10:51TED Talk
- 14:07How Jennifer Got Started
- 16:31The Procedure
- 17:31Annotating
- 21:51Data
- 26:56Initial Questions
- 28:51Documentation
- 29:18Sensor Quality
- 30:15Mapping
- 31:00Research
- 33:59Azure ML
- 35:53Supervised ML
- 47:45Resources
- 48:59Prototype
- 52:47Offsite test
- 54:07Calibration
- 57:53Modifications
related talks
01h 01mKeynote: Using EEG and Machine Learning to Perform Lie Detection
31mJennifer Marsman, Principal Developer Evangelist, Microsoft
26mBrainDrain Using Machine Learning and Brain Waves to Detect Errors in Human
06mExploration of Person-Independent BCIs for Internal and External Attention-Detection in ...
29mAI & Neurobiology of Deception: How AI Helps Us Understand Ourselves
11mConnected brains
32mIrina Rish, Research Staff, IBM T.J. Watson Research Center
22mA Classification Model for Sensing Human Trust in Machines Using EEG and GSR