CustomerRevenuePrediction

In many businesses, identifying which customers will make a purchase (and when) and how much will they spend, is a critical exercise. This is true for both brick-and-mortar outlets and online stores. This project's data is website traffic data acquired from an online retailer.

Data URL: Kaggle Link

The challenge: Predict total sales

The data provides information on customer's website site visit behavior. Customers may visit the store multiple times, on multiple days, with or without making a purchase. The variable $revenue$ lists the amount of money that a customer spends on a given visit in the dataset. My main goal for this project is to predict how much money a customer will spend, in total, across all visits to the site, during the allotted one-year time frame (August 2016 to August 2017).

Prediction target

More specifically, I am predicting the transformation of the aggregate customer-level sales value based on the natural log. That is, if a customer has multiple revenue transactions, then the sum of all the revenue generated across all of the transactions, i.e.,:

$$ custRevenue_i = \sum_{j=1}^{k_i} revenue_{ij} \ \ \ \forall i \in customers $$

$$ \text{where } k_i \text{ denotes the number of revenue transactions for customer } i $$

And then transform this variable as follows:

$$ targetRevenue_i= \ln(custRevenue_i + 1) \ \ \ \forall i \in customers $$

Modelling

For this project, I have used $Linear Discriminant Analysis$ and $Multivariate Adaptive Regression Splines$ to predict the following:

1. $LDA$ for classifying the customer if they will buy something or not.
2. $MARS$ for predicting how much they might spend, in terms of $logarithmic$ value.