Activity Detection Using KNN

This project aims to predict physical activities based on sensor data using a K-Nearest Neighbors (KNN) model. The sensor data consists of 3-axis accelerometer readings from wearable devices such as sensors placed on the thigh and back. These readings are used to predict various physical activities like walking, running, sitting, or standing.

The application is built using Streamlit for the web interface, allowing users to either upload data or manually input sensor values for real-time predictions. The model predicts the activity based on the input and displays an activity label along with a related GIF.

Features:

Manual Data Entry: Users can enter accelerometer data from 6 features (X, Y, Z axes for both back and thigh sensors) and receive an activity prediction.
File Upload: Users can upload a CSV file with sensor data, and the system will predict the activities based on that data.
Prediction Output: For each activity prediction, the app displays a label and a related GIF for better visualization.

Setup and Usage

Clone the Repository:

   git clone https://github.com/visionjain/HARTH-Activity-Detection
   cd HARTH-Activity-Detection

Install Dependencies:

Install all required dependencies from requirements.txt:

   pip install -r requirements.txt

Run the Streamlit App:

After the dependencies are installed, you can start the app using:

   streamlit run app.py

Using the App:

Manual Data Entry: Enter values for the accelerometer data in the input fields for each of the 6 features. After entering the data, click the "Predict Activity" button to get the predicted activity and a related GIF.
Upload CSV: If you have a dataset in CSV format with sensor data, you can upload it to get predictions for each row.

Data Format

For the file upload functionality, the CSV should contain the following columns:

feature_1: Back sensor X-axis acceleration (g)
feature_2: Back sensor Y-axis acceleration (g)
feature_3: Back sensor Z-axis acceleration (g)
feature_4: Thigh sensor X-axis acceleration (g)
feature_5: Thigh sensor Y-axis acceleration (g)
feature_6: Thigh sensor Z-axis acceleration (g)

Model Description

The core of the prediction system is a K-Nearest Neighbors (KNN) model that is trained using data from a Human Activity Recognition (HAR) dataset. The model uses 6 features (3 for back sensor and 3 for thigh sensor) to predict the type of activity the user is performing.

The KNN model works by comparing the input feature vector with the stored labeled data and finding the most common activity in the nearest neighbors.

Activity Labels:

1.0: Walking
2.0: Running
3.0: Shuffling
4.0: Stairs (ascending)
5.0: Stairs (descending)
6.0: Standing
7.0: Sitting
8.0: Lying
13.0: Cycling (sit)
14.0: Cycling (stand)
130.0: Cycling (sit, inactive)
140.0: Cycling (stand, inactive)

Demo

Contributing

Contributions are welcome! Feel free to fork the repository and submit pull requests for any improvements or bug fixes. If you find any issues or have suggestions for new features, please create an issue in the repository.