AtmosafeAI

Using Machine learning based regression models to predict the pollution levels, and the PM2.5 levels in order to help users think if they need to wear a mask while going out or not, for New Delhi.

Comment

Inspiration

Since my birth, I have primarily resided in Delhi, which is frequently referred to as the pollution capital of the globe. I've often pondered how we can make this attractive and prosperous city more inhabitable and assist the populace in combating this life-threatening issue. It motivated me to discover how pollution is measured and how we can contribute to its mitigation in a responsible and sustainable manner.

What it does

It helps to make the user decide whether or not they need to wear a mask based on the prevailing Pollution Conditions, for which it is manually inputted at the moment, but I am working on implementing a custom API to get the data from the government of India website for the same

How we built it

I built it using SwiftUI and Xcode. I first made the basic layout for the application, and then simultaneously worked on the Machine Learning model in Python using Tensorflow, which was converted to be used in a Swift application. I also integrated the use of SceneKit to display beautiful low poly 3D models, which are interactive with infinite degrees of freedom and provide more interactivity with the user. I finally used the Charts API just launched by Apple last year to display beautiful charts and provide users a range of data give them more insight on the typical trailing values of the data

Challenges we ran into

I had a huge problem training the ML model, and testing it on real world data. I first tried with multiple random forest models, with 100 iterations and had quite a lot of issues with the accuracy. I then ended up with using two one hot encoders, tabular regression and Boosted Regression to get an extremely low root mean square error and minimum absolute percentage errors, something which was viable in production.

Accomplishments that we're proud of

I'm extremely proud of the way UI turned out for the application and how intuitive it has been to friends on whom I tested the application. I also love the fact that I've been able to use Machine Learning in a real world application and have it develop and give practical results for the same.

What we learned

I've learned how to use UIKit to help display confetti animation. I also learned how to display smooth, intuitive and good animation.

What's next for AtmosafeAI

I'd like to train more Machine Learning models based on global data to provide the application to more and more users. Additionally I'd love to integrate API access with the same, and have it detect based on the users location and notify about the same, and look into using mobile devices like Apple Watch as well.

Built With

Share this project:

Updates

Submission history