-
Azure Machine Learning
-
Close look on Azure Machine Learning
-
Current Location with pins of recent incidents
-
Notification being sent when inside a dangerous zone
Inspiration
Sexual Harassment has always been a topic that people have been talking about but it is not something that we can easily eliminate or prevent in a short period amount of time. We thought it would be cool and useful if people can know how likely that a sexual harassment event is going to happen to themselves based on the time and location. With that information, people can stay more alert and better protect themselves from predators!
What it does
The app is going to keep track of the user location and current time. Then the app will look up the database that has the prediction value (using Microsoft Azure) of the probability of how likely that a sexual harassment event is going to happen at this location and this specific time. If the probability pass the default threshold that we set, the app will send out a notification to alert the user to be aware of the surroundings/be careful even if the user is not currently using the app! The app also displays similar crimes that happen near the user's location.
How we built it
We got the LAPD crime data through the government website then we used R to clean the and extract certain amount of data that we believed is critical for the machine learning experience. We also executed some R and python scripts inside Azure Machine Learning Studio to help us better understand which data might affect the prediction outcomes. We tried different algorithms such as poisson regression, linear regression and boosted decision tree regression to help predict the most accurate probability of the next sexual harassment event. We trained our model based on the occurrences of past incidents and we are trying to predict the amount of incidents in the future and divide that by average amount to get the probability and output it to database. We use android studio and Google Map API to keep track of user location and pin point location of assaults or similar crimes.
Challenges we ran into
We were struggling on which algorithm to use or whether we should develop our own model that might fit the distribution better because we are using huge amount of raw data directly from LAPD. We finally decided to use boosted decision tree regression but the whole process took a very long time. Aside from that, we were also facing challenges of not knowing the full data such as the victim's personal information. We were also not sure how to pre process the data and clean it before passing it into any machine learning algorithm. Another challenge that we faced was being able to track user location and display nearby assault incidents in order to make users more alert and better protect themselves from predators.
Accomplishments that we're proud of
Working on environment and technologies that we are not familiar with. Working 36 hours intensely for a cause that we believe in, a cause that can potentially decrease crime rate and increase people's safety.
What we learned
Azure Machine Learning Studio. Azure Deployment. Google Maps API.
What's next for StayAlert
Better optimize Machine Learning algorithm. Maybe introduce neural networks to assign different weights based on different factors such as incident dates, gender and age to help improve the accuracy of the prediction. We can also add another feature where users can search a specific location through the app and get a better understanding around that location.
Log in or sign up for Devpost to join the conversation.