Activity Detection Model

A deep learning model for detecting worker productivity states using accelerometer data. This model classifies worker activities into four categories: value-adding work, non-value-adding work, walking, and idling.

Overview

The productivity state detection module identifies worker productivity states based on accelerometer data from wearable devices. The states are defined as follows:

• Value-adding work: Activities directly leading to task completion (e.g., painting a wall)
• Non-value-adding work: Activities indirectly supporting task completion (e.g., mixing paint)
• Walking: Worker movement/travel
• Idling: Non-working states (e.g., resting, talking)

Project Structure

Activity-Detection-Model/
├── data/                    # Dataset directory
│   └── labelled_dataset.csv # Input dataset
├── src/                     # Source code
│   ├── config/             # Configuration module
│   │   ├── __init__.py
│   │   └── config.py       # Model and training parameters
│   ├── data/               # Data handling module
│   │   ├── __init__.py
│   │   └── data_loader.py  # Data preprocessing and loading
│   ├── models/             # Model architecture module
│   │   ├── __init__.py
│   │   └── cnn_model.py    # CNN model implementation
│   └── train.py            # Main training script
├── requirements.txt         # Project dependencies
└── README.md               # Project documentation

Features

• Real-time activity classification using accelerometer data
• CNN-based deep learning model
• Automatic data preprocessing and balancing
• Early stopping to prevent overfitting
• Model checkpointing for best model preservation
• Comprehensive evaluation metrics and visualizations

Installation

1. Clone the repository:

git clone https://github.com/yourusername/Activity-Detection-Model.git
cd Activity-Detection-Model

2. Install dependencies:

pip install -r requirements.txt

Usage

1. Place your dataset file (labelled_dataset.csv) in the data/ directory. The dataset should contain the following columns:

   • timestamp
   • acc_x_h, acc_y_h, acc_z_h (head accelerometer data)
   • acc_x_w, acc_y_w, acc_z_w (waist accelerometer data)
   • acc_x_c, acc_y_c, acc_z_c (chest accelerometer data)
   • activity (activity label)

2. Run the training script:

python src/train.py

The script will:

• Load and preprocess the data
• Train the CNN model
• Generate training history plots
• Create a confusion matrix
• Save the trained model and scaler

Model Architecture

The model uses a CNN architecture with:

• Three convolutional layers (32 filters each)
• Dropout layers (0.2 rate)
• Two dense layers (64 and 4 units)
• Softmax activation for classification