CNN + LSTM for Activity Recognition

Overview

This project utilizes a hybrid CNN + LSTM architecture for activity recognition using time-series sensor data from the ARAS B House dataset. The model integrates Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies.

Model Architecture

The CNN + LSTM architecture consists of:

• Input Layer: Defines the input shape of time-series data.
• CNN Layers:
  • Conv1D Layer: Extracts spatial features from input data.
  • MaxPooling1D Layer: Reduces dimensionality while preserving critical features.
• LSTM Layers:
  • LSTM Layer: Captures temporal dependencies in sequential data.
  • Dropout Layer: Prevents overfitting by randomly deactivating neurons.
• Fully Connected Dense Layer: Processes extracted features for classification.
• Output Layers:
  • Two softmax layers to classify activities for Resident 1 and Resident 2.

Key Features

• CNN for spatial feature extraction: Identifies localized patterns in sensor data.
• LSTM for temporal learning: Recognizes activity sequences over time.
• Softmax output layers: Enables multi-class classification.
• Dropout and regularization: Reduces overfitting and enhances generalization.

Performance

Metric	Resident 1	Resident 2
Accuracy	94.73%	93.44%
Precision	94.59%	94.05%
Recall	95.05%	96.06%
F1 Score	94.66%	94.79%

Installation

Prerequisites

Ensure you have the following installed:

• Python 3.x
• TensorFlow (pip install tensorflow)
• NumPy (pip install numpy)
• Pandas (pip install pandas)

Steps

1. Clone the repository:

   git clone https://github.com/yourusername/CNN-LSTM-ActivityRecognition.git
   cd CNN-LSTM-ActivityRecognition

2. Install dependencies:

   pip install -r requirements.txt

3. Run the training script:

   python train.py