GAIT AUTH
SMARTWATCH GAIT AUTHENTICATION
MATLAB machine learning project for continuous user authentication using smartwatch motion sensor data.
YEAR
2025
ROLE
ML Engineer
TYPE
AI/ML
ORIGIN
UNIVERSITY
STACK
MATLAB · Neural Networks · Machine Learning
06(OVERVIEW)
A machine learning project using smartwatch motion data to continuously authenticate users via walking patterns. Extracts statistical features from accelerometer and gyroscope signals and uses neural networks for user recognition.
THE PROBLEM
Traditional PINs only verify at login. Wearables handling personal data need continuous, passive authentication to verify the active user without interrupting usage.
THE SOLUTION
We built a pipeline using smartwatch motion sensors. Raw signals are preprocessed, segmented, and converted to feature vectors. Custom FFMLP neural networks identify users by unique motion signatures.
THE OUTCOME
Achieved 93.38% accuracy and 2.19% EER across ten users. Demonstrated reliable continuous authentication on wearables using motion features and simple neural networks.
(KEY FEATURES)
- 01Signal preprocessing & 10-second window segmentation
- 02Extraction of 36 statistical gait features
- 03Custom user-wise one-vs-all neural network models
- 04Multi-class prediction with probability scores
- 05Performance metrics: Accuracy, FAR, FRR, EER
(NEXT PROJECT — 07)