Inspiration
Prize money(Duh!)
Storyline
A natural calamity has just hit the east coast. The devastation is inexpressible and the response teams have been on high alert. Huge number of victims have been requesting help online directed towards friends and authorities. A team from PSU brought a “smart mobilization system” to aid the response teams. All the users sending the emergency appeals are now channeled through a central App and a “cool” NLP module is helping the teams to direct the required help towards the victims. All without any hassle. Furthermore, the team has helped the ERU (Emergency Response Units) with a wildfire in no time. No wonder, the team has won many hearts and definitely won the HackPSU.
What it does
It enables a richer communication between the rescuers and the victims at risk to identify the immediate aid services required. An additional feature we added is the detection of forest fires and alert systems to the people in the area.
How we built it
Method - Message Annotation Cleaned the messages for removing numbers and special characters. Tokenized the messages and the stop words are removed. Further cleaning to split the hashtags into a complete sentence. Processing to get full sense of known abbreviations. Labelled data trained with a GridSearchCV model. Any new emergency request will now be annotated with appropriate tags.
Method - Disaster Prediction Limiting the disaster to Wildfires for the scope of HackPSU. Labelled Images are trained on a CNN. Binary Cross Entropy Loss and RMSProp. Polling satellite data periodically to make predictions based on the learned features. Responding to the prediction by broadcasting safety alerts to all the citizens in the vulnerable areas.
LAMP Stack Hosted on a GCP VM instance running Ubuntu Apache for web server MySQL database for all the citizen, appeal data and annotations. PHP as a backend with integration of machine learning with Python
Challenges we ran into
Limited clean datasets. Hyper parameter tuning to build a top-class model. Sleepless night.
Accomplishments that we're proud of
A problem worth solving as at the times of calamity, in a chaos, even the best of response teams are bound to human error in panic. A smart system like ours will be working hand-in-hand to aid the response teams for better decisions. Help reached to a victim at right time can save their life. We are sure it's priceless.
What we learned
Learnt using Lamp stack. Building models using deep learning. Twilio API for alert SMS.
What's next for Smart mobilization system
Use the data from other geospatial data sources and build a robust model for other calamities. Have a feedback system that works between the citizens and rescue organizations. Model to learn the address from the emergency appeals made by the citizens.
Log in or sign up for Devpost to join the conversation.