AccelAuth ML
Acceleration-Based Continuous Authentication
A neural-network authentication system that verifies user identity from acceleration behavior, designed as a stronger alternative to static password-only access.
Primary audience
Story
Why this project stands out
The problem
Passwords and PINs are easy to steal, reuse, and bypass.
Modern devices need transparent and continuous verification that does not interrupt users every few minutes.
The approach
Model user movement patterns with an FFMLP neural network.
Used acceleration-based features from multiple users, then optimized model quality through feature reduction and hyperparameter search.
The outcome
A viable behavioral-authentication baseline with strong classification metrics.
Results showed high precision and recall trends, with optimization improving stability and reducing misclassification rates.
Overview
About the Project
AccelAuth ML evaluates continuous user authentication using acceleration-derived behavioral features. The project uses a Feedforward Multi-Layer Perceptron (FFMLP) model with time-domain and frequency-domain inputs to identify users and detect impostor behavior. To improve robustness, the pipeline applies optimization methods such as PCA, recursive feature elimination, and systematic hyperparameter tuning with cross-validation.
Capabilities
Key Features
Behavioral Authentication
Classifies users using acceleration behavior instead of relying only on static credentials like passwords or PINs.
FFMLP Neural Network
Implemented a feedforward multi-layer perceptron with tuned hidden layers, activation setup, learning rate, and epochs.
Time and Frequency Features
Used a feature space combining temporal and spectral acceleration characteristics for stronger user differentiation.
Feature Optimization
Applied PCA and recursive feature elimination to reduce noise and focus training on the most informative dimensions.
Hyperparameter Tuning
Used grid search and k-fold cross validation to select parameter combinations with better generalization behavior.
Explainable Evaluation
Reported confusion matrix breakdowns and inter/intra-user variance statistics to interpret reliability and error patterns.
Gallery
Screenshots
Engineering
Challenges & Solutions
Results
Project Outcomes
Delivered a complete acceleration-based authentication ML pipeline
Demonstrated strong viability for continuous and transparent user verification
Improved baseline performance through feature and hyperparameter optimization
Produced clear statistical reporting for model behavior and misclassification analysis
Established a foundation for larger-scale biometric security experimentation