3Y1B | Devpost

Neural Network Data over 10 Epochs
Loss vs # Epochs and F1 Score vs #Epochs
3Y1B Logo

Inspiration

We were inspired by Amazon and their ability to retain customers. Amazon has amazing customer service and they often communicate with their customers in the form of recurrent advertisements online and through email. They also have a membership program with many perks like insanely fast shipping.

What it does

We preprocessed the dataset and ran many ML models including Logistic Regression, KNNs, Adaboost, Gradient Boosting, XGBoost, Decision Trees, Random Forest, and most notably, a Neural Network. These models would predict whether the customer datapoint will churn or not churn depending on the data features.

How we built it

We used pandas and numpy to preprocess the data. We used one-hot encoding to turn the categorical data into numerical data, and we also oversampled the minority class and filled in missing data. Then, we used sklearn to run the classical ML models and PyTorch to implement a Neural Network.

Challenges we ran into

We faced some challenges collaborating. When we used the same Google Colab Notebook, sometimes things would not be synced, or we would have permission issues. In terms of coding, we all had our minor issues with implementation, but we were able to overcome these issues through debugging.

Accomplishments that we're proud of

We are proud of achieving a F1 score of around 98% using the Neural Network on the testing dataset.

What we learned

We learned how to go through the whole process of analyzing a dataset, going from initial exploration to preprocessing to model implementation and final analysis. For example, we learned about numerous pre-processing techniques to turn categorical data into numerical data and the different code a Neural Network.