-
-
-
-
-
-
Azure ML studio model training and evaluation
-
Azure functions deployed endpoint
-
Linode
-
Linode
-
Inspiration
Multilevel threads have increased every day for the safety and security team. Unfortunately, we noticed that the crime information today if it is either not available, is very limited and difficult to access by the public. At the end of the day, perpetrators have always benefited from this lack of information.
What it does
So that’s how we decided to build Cyber.eye, an app to democratize public safety-related information around the world. Each user can also report different types of criminal acts on our platform, and by using Azure Machine Learning Services, we will combine the data and predict & prevent crime from even happening.
Users can define areas they visit regularly (home, work, their children’s school) and they’ll receive real-time alerts about everything that is happening in those areas. We are making all of this information public and eventually providing this information to the authorities, which helps authorities, the police, to distribute their limited resources more efficiently.
How we can start constructing that path, is to decentralize this kind of safety-related information by using blockchain and open up this path by allowing people to report everything that happens. By actively participating in Cyber.eye, you will have valuable prevention tools to help your family, friends from crimes and hopefully, save the lives of many people.
Technical Features:
- Linode hosted functions for cockroach DB cloud access as well as blockchain services
- Azure serverless functions to serve safety score ML model
- ML model is trained on data obtained from the FBI NIBRS dataset from 2015 - 2020 using Azure ML and Azure Cognitive Services
- Azure ML studio was used to load in various crime datasets from the FBI NIBRS data
- Azure Cognitive Service Anomaly detector was used to detect anomalous crime data or incidents
- On Azure ML studio, we experimented with various models and evaluated accuracy
- We then used Azure ML notebooks for some custom model training as well
- Lasso regression was optimal for our purpose but this can be improved
- The final model was pickled and saved, then deployed as an endpoint on serverless Azure Functions
- All data is location-specific and limited to US cities and regions currently
- Crowdsourced incident verification is built as a feature of the mining mechanism on the blockchain service
Features
- Probability of crime by type
- Probability of racially motivated crimes
- AI determined neighborhood score
- Mass notification of the crime (reporting feature)
- Aggregate score depending on the mode of travel
- Show police presence / recent crimes
- Heatmap of crime sourced from gun crime and FBI databases (via the Gaussian Kernel Density Estimation Algorithm)
- Notes from crime reports
- Decentralized crime reporting on blockchain
- Set alerts for up to date notifications on recent crime in selected areas
Challenges we ran into
Large datasets and incremental loads mean we had to deal with scalability issues. We didn't want responsiveness to be sluggish so we had to add a lot of caching, compression, data splitting, and data reshaping to ensure an ideal experience.
Accomplishments that we're proud of
Deeply interesting data visualized in an easily accessible way. We found it very interesting interacting with the various data points we had plotted on the map. We are also extremely proud of leveraging Azure ML and Azure Cognitive Services in conjunction with Azure Functions to derive amazing insights about the safety of different regions across the United States.
What we learned
First time using Deck.GL for our mapping and Linode to host our blockchain.
What's next for Cyber.Eye
- Packway more data into our dataset displayed and our ML model
- Allow filtering by crime type, locale, and time
- Add cleanliness and 311 data reporting
Software Architecture
Log in or sign up for Devpost to join the conversation.