Viettel Big Data - Churn Prediction (PENtagon Team)

Inspiration

Viettel gave us a 1-million-row dataset from their own customer data, which is a great oppoturnity for students like us to practice machine learning with (pretty) big data.

What it does

Given a piece of Viettel (telco) customer data, our model can intelligently tell if he is going to cancel service in the near future (next few months) or not.

How we built it

The tool we use is RStudio, a nice and free tool tailored for data scientists. The science behind? It is deep learning, an emerging field in AI. We built a multilayer perceptron (MLP) with 5 layers in the keras framework (in R). The model was trained on the given dataset.

Challenges we ran into

We are new to R (started to use it in this contest).

Accomplishments that we're proud of

We are able to learn R and deep learning fast. We are able to think of a way to transform the given data.

What we learned

R, MLP, keras, recipes...

What's next for Viettel Big Data - Churn Prediction (PENtagon Team)

We're not stopping. It's our goal now to dive deeper into the already deep learning world ^^

How to use our model?

In the submitted zip file, there is a R file called predict.R and a model file called pp_relu64_relu64_sig.

• Change the input file path from E:/Study/Hackathon/bai1test/test2 to the path of the csv file used for testing.
• Change the path to the model file from E:/Study/Hackathon/pp_relu64_relu64_sig to the correct path on the testing platform.

Run the file to see a truth table of the prediction result.

Built With

• keras
• r