As a part of a HSBC Hackathon we created a simple feedforward neaural network to predict customer behavior
Summary
For the hackathon HSBC gave us 2gbs of text files containing data on their customer transaction records, products purchased, debt info, etc. Using this data I was to develop an algorithm that predicted if a new customer would purchase a HSBC prodcut within the next 3 months. A lot of this project was simply preparing the data to be able to put into a neural network because the data we were given was very unstructured and contained both numerical and categorical data. I made the categorical data points into one-hot vectors and then fed them along with the numerical data points into a simple perceptron algorithm which gave us very good results in loss and accuracy.
The Pre-processing required to get data in format where I could input it into the neural network
Graph of how training & validation loss fell and accuracy rose over the iterations