Human Activity Recognition using Smartphone

Introduction:-

This project senses human activity with the help of an app installed in the smartphone which uses ML to predict from the data collected from sensors present in the phone. This project is handled by the DSC ML Team of NIT Patna

Which Sensors we use to get the data:-

1. Accelerometer

2. Gyroscope

How it works?

Once we collect the data we will be preprocessing the data and train it using LSTM or further more advanced models. Then finally we would be predicting the activity.

Activities It Tracks:-

1. WALKING
2. WALKING_UPSTAIRS
3. WALKING_DOWNSTAIRS
4. SITTING
5. STANDING
6. LAYING

Tasks Involved:-

1. Firstly we need to make a android app and integrate with firebase for collecting the data
2. Then we need to preprocess the data
3. Create a ML Model
4. Need to deploy the model in Mobile
5. Make it avalaible in Playstore

Approaches Used:-

We used 2 Different Strategies in order to achieve the accuracy. They are

1. We took each frame of the data into 128 steps and created 561 different possible calculations and passed them using machine learning algorithms like Logistic Regression ,KNN ,Decision Trees ,Random Forest ,Support Vector Machine etc.

   The below is the metrics which are resulted from above method.

Model Name	Accuracy Acheived
Logistic Regression	98.66%
Support Vector Machine	93.42%
XGBoost	98.73%
Linear SVM	96%
Decision Tree	92.5%
Random Forest	92.5%
KNN	91%

2. We took each frame of data and passed it to the recurrent neural network i.e LSTM and predicted the activity it belongs to.

   The below is the metrics which are resulted from above department.

Model Name	Accuracy
Stacked LSTM	91.78%

Maintainers of this Project:-

• Sai Durga Kamesh Kota
• Raj Kothari
• Aparna Juhi