Deep stacked residual bidirectional LSTMs for HAR
A deep learning architecture using stacked residual bidirectional LSTM cells with TensorFlow for human activity recognition from sensor data.
What is Deep stacked residual bidirectional LSTMs for HAR?
HAR-stacked-residual-bidir-LSTM is a deep learning research project that implements a neural network architecture using stacked residual bidirectional Long Short-Term Memory (LSTM) cells for Human Activity Recognition. It classifies human movements from wearable sensor data into categories like walking, sitting, or standing, achieving improved accuracy over previous methods. The architecture is designed to be easily configurable for different datasets with fixed time windows.
Machine learning researchers and practitioners working on time-series classification, particularly those focused on human activity recognition from sensor data or similar sequential data problems.
Developers choose this project for its innovative combination of residual connections with bidirectional LSTMs, which enables deeper networks and better gradient flow while maintaining competitive accuracy on standard HAR benchmarks. The configurable architecture allows easy experimentation with different network depths and dataset adaptations.
Overview
Using deep stacked residual bidirectional LSTM cells (RNN) with TensorFlow, we do Human Activity Recognition (HAR). Classifying the type of movement amongst 6 categories or 18 categories on 2 different datasets.
Use Cases
Best For
- Classifying human activities from smartphone or wearable sensor data
- Research on deep LSTM architectures with residual connections
- Time-series classification tasks with fixed-length windows
- Experimenting with bidirectional RNN variations
- Adapting neural networks to new sensor-based datasets
- Comparing performance against state-of-the-art HAR models like DeepConvLSTM
Not Ideal For
- Teams using modern deep learning frameworks like TensorFlow 2 or PyTorch
- Projects requiring real-time or streaming sensor data processing
- Applications where state-of-the-art accuracy is critical, such as competitive benchmarking
- Developers seeking production-ready code with active maintenance and extensive documentation
Pros & Cons
Pros
Combines residual connections with bidirectional LSTMs to enable deeper networks and improve gradient flow, achieving 94% test accuracy on the HAR dataset as documented.
Allows easy adjustment of network breadth, depth, and sequence length through configuration files, facilitating experimentation with different datasets like HAR and Opportunity.
Designed for fixed-window sensor data, making it straightforward to adapt to new HAR datasets with minimal code modifications, as shown in the dual dataset support.
Cons
Built on TensorFlow 0.11 and Python 2, which are deprecated and unsupported, complicating setup and integration with modern tools.
The README notes a lower F1-score (0.893) compared to DeepConvLSTM (0.9157) on the Opportunity dataset, indicating it isn't the top performer for all benchmarks.
Requires specific data handling with dependencies like scipy and pandas, and fixed-window segmentation, which can be cumbersome and error-prone for new users.
Frequently Asked Questions
Related Projects
Tensorflow - Open source software library for numerical computation using data flow graphsAn Open Source Machine Learning Framework for Everyone
Classification on raw time series in TensorFlow with a LSTM RNNHuman Activity Recognition example using TensorFlow on smartphone sensors dataset and an LSTM RNN. Classifying the type of movement amongst six activity categories - Guillaume Chevalier
skflowSimplified interface for TensorFlow (mimicking Scikit Learn) for Deep Learning
Predict time seriesSignal forecasting with a Sequence-to-Sequence (seq2seq) Recurrent Neural Network (RNN) model in TensorFlow - Guillaume Chevalier
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.