Selecting and Filtering Data
Showing Data for Texas
Showing Data for Illinois
We wanted to enlighten Americans on future gun violence events to help guide gun policies and assist law enforcement. Gun violence is a prominent issue in America, with 260,000 gun-related crimes in the last 6 years. While there is still much debate about gun policies, we decided to find a simple way to communicate the facts of recent gun-violence events in the USA.
What it does
Our application uses machine learning to predict the probability of future gun violence events throughout America for the next three years based on hundreds of thousands of data points from past events. You can visualize the number of gun-related crimes on a US choropleth map filtered by a date-range of your choosing. You can also select dates up to 2 years in the future, which predicts the number of gun-events expected to occur in each state. There is also an analysis of the top 3 counties in each state that face the highest number of gun-related crimes to help guide law enforcement in their effort to bring peace to struggling areas.
How we built it
The backend was built using python, keras, and scikit-learn for the time-series machine learning models, while also using data-manipulation tools such as numpy and pandas for feature engineering and efficient processing of the data points. The front end was also built with python, using plotly for our heat map and Flask to run our code locally.
Challenges we ran into
A simple machine learning model would not work with the dataset because it consisted of time-series data, so it took some time to understand how to manipulate the data and the machine learning models to make future predictions based on the timing of past events. Evaluating the quality of our predictions was also a challenge due to the nature of the problem of predicting future occurrences in such a large scope. For the frontend, there were some struggles with cleaning the input CSV files so that our visualization would allow different inputs depending on the problem the user would like to visualize.
Accomplishments that we're proud of
We are proud that we were able to complete our product from the ground up in 12 hours. Not only were we able to setup a successful machine learning model, but we also found an aesthetic way to share our data and findings with the user through the web app. We also learned much about python, machine-learning, and data visualization during the hackathon.
What's next for MarcoGunPolo
We hope to expand our visualization to allow for more detailed analysis of state and county data related to gun violence. We also plan to expand our model to make use of more of the feature data to further enhance our visualization and provide more concrete, localized details about this huge issue in our country.