guillaume-chevalier/LSTM-Human-Activity-Recognition

# Technical Summary Employs a many-to-one LSTM architecture that processes 128-sample time windows of 9-channel inertial sensor data (3-axis accelerometer and gyroscope readings) without extensive feature engineering, relying instead on the recurrent network to automatically learn temporal patterns across sequential measurements. Minimal preprocessing is applied beyond gravity filtering, contrasting with traditional signal-processing-heavy approaches that require manual feature extraction. Built with TensorFlow and includes Jupyter notebook implementations demonstrating end-to-end data loading, model training, and evaluation metrics on the UCI HAR Dataset.

3,549 stars. No commits in the last 6 months.