Connect Pulse

Inspiration

Knowing the pain of lagging in a multiplayer game, we want to help people get their internet fixed before they realize it. Instead of "rubberbanding" in game, we want to sling people back from the depths of high ping into seamless and fun gaming experiences.

What it does

Given customer data from Frontier dataset, we predict if the customer may need a service repair or not.

How we built it

Jupyter Notebook (numpy, pandas, tensorflow, keras, sklearn, matplotlib)

Challenges we ran into

In regard to the deep neural network, we ran into various problems with deciding which predictor to use and overfitting. We also try to start with the LSTM model, but the multivariate and multi-step variables require a significant modification to the input data to be usable for the LSTM model.

Accomplishments that we're proud of

Making our first deep learning neural network!

What we learned

We learned how to preprocess data (filling in missing values, dropping columns, one-hot encoding), shape data into appropriate size, analyze data to good predictor variables, create a model, train the model, and use the model to predict test data. We also learn about LSTM and its variation, multivariate and multi-step.