Human Activity Recognition (HAR) using Deep Learning

Foundations of Data Science (FDS) 2025/2026 Sapienza University of Rome

Project Overview

This project focuses on Human Activity Recognition (HAR) based on time-series data collected from smartphone sensors (accelerometer and gyroscope). The goal is to classify various physical activities (e.g., Walking, Standing, Laying) using Deep Learning techniques.

The project utilizes the UCI HAR Dataset and implements two distinct neural network architectures to compare performance and training strategies:

1. CNN-LSTM: A hybrid end-to-end model combining Convolutional Neural Networks (for feature extraction) and LSTMs (for temporal sequence modeling).
2. ConvAE-LSTM: A two-stage approach involving a Convolutional Autoencoder (pre-trained for feature compression) followed by an LSTM Classifier (transfer learning).

Project Structure

Actify/
│
├── cfg/                          # Hydra configuration files
│   ├── base.yaml                 # Base configuration
│   └── net/
│       ├── cnn_lstm.yaml         # CNN-LSTM model configuration
│       └── convae_lstm.yaml      # ConvAE + LSTM configuration
│
├── data/                         # Dataset directory 
│
├── src/
│   ├── dataset/
│   │   ├── dataset.py            # Dataset download
│   │   └── loader.py             # Data loading utilities
│   │
│   └── net/
│       ├── blocks/               # Reusable network building blocks
│       │   ├── cnn.py
│       │   ├── conv_deconv.py
│       │   └── lstm.py
│       │
│       └── models/               # Model definitions
│           ├── cnn_lstm.py        # CNN-LSTM model
│           ├── convae_lstm.py     # ConvAE + LSTM model
│           └── __init__.py
│
├── experiment.py                 # Main training & evaluation pipeline
│
├── sweep_cnn_lstm.yaml            # W&B sweep configuration (CNN-LSTM)
├── sweep_convae_lstm.yaml         # W&B sweep configuration (ConvAE-LSTM)
│
├── README.md                     # Project documentation
├── pyproject.toml                # Project dependencies and configuration
├── uv.lock                       # Dependency lock file
├── .python-version               # Python version
└── .gitignore

How to Run the Project

This project uses uv as the Python package manager and Hydra for configuration management.

1. Clone the repository and enter the project folder:

   !git clone https://github.com/Catairu/Actify.git

   !cd Actify

2. Install the dependencies using uv:

   !uv sync

3. Download and preprocess the UCI HAR Dataset:

   !uv run python src/dataset/dataset.py

4. Select the model to run by editing the configuration file: cfg/base.yaml

   Set the net name field to one of the following values:

   • cnn_lstm : CNN-LSTM end-to-end architecture
   • convae_lstm : Convolutional Autoencoder + LSTM

   Example: name: cnn_lstm or name: convae_lstm

5. Run training and evaluation:

   !uv run python src/experiment.py

Evaluation metrics:

• Accuracy
• F1-score
• Confusion Matrix

For further details on the methodology, experiments, and results, please refer to the project report.