Loan Default Neural Network

Aim

To demonstrate the implementation of a multi-layer perceptron (MLP) neural classifier model with one hidden layer.

Dataset

The data was sourced from www.lendingclub.com. The complete loan dataset was used from 2007-2011.

Methodology

1. The data was cleaned to exclude all columns that were deemed unnecessary for the purpose of the build, for example the notes/comments column. Primarily numerical and categorical data was used for the prediction.
2. Convert all categorical variables to boolean ones using a pandas function called get_dummies
3. Split the dataset into test and train
4. Normalise all features as the MLP model used is particularly sensitive to this.
5. Run the model. A MLP classifier is used with default values.
6. Run model over test dataset
7. Test the accuracy of the model

Outcome

Simply using a single hidden layer of 3 perceptrons, a 99% accuracy was achieved.