Udacity Data Scientist Nanodegree

Capstone Project: Arvato Financial Solutions

This repository contains files associated with my capstone project submission for the Udacity Data Scientist Nanodegree. The blog post that accompanies this can be found on Medium.

Table of Contents

1. Installation
2. Project Motivation
3. File Descriptions
4. Results
5. Licensing, Authors, and Acknowledgements

Installation

This project was performed using Python 3 and the following libraries:

• numpy
• pandas
• matplotlib
• sklearn
• IPython

Project Motivation

The objective of the project is to predict whether the recipient of an advertising campaign is likely to respond to it, and comprises three major steps:

1. Customer Segmentation Report
   Unsupervised learning methods (PCA, k-means clustering) will be used to analyse attributes of established customers and the general population in order to create customer segments.

2. Supervised Learning Model
   Using a third dataset with attributes from targets of a mail order campaign, the previous analysis will be used to build a machine learning model that predicts whether or not each individual will respond to the campaign. This model will be optimised via hyperparameter tuning and other methods.

3. Kaggle Competition
   With the optimised model, predictions will be made on the campaign data and individuals will be ranked by how likely they are to convert to being a customer. The predictions will be submitted to a Kaggle competition.

File Descriptions

The following files are included in this GitHub repository:

• Arvato Project Workbook.ipynb: Jupyter Notebook containing the data analysis
• Arvato Project Workbook.html: HTML version of the above notebook
• README.md: this file
• kaggle_screenshot.png: screenshot of result from submission to Kaggle competition

The following data files are NOT included, but are required to run the notebook:

• Udacity_AZDIAS_052018.csv
• Udacity_CUSTOMERS_052018.csv
• Udacity_MAILOUT_052018_TEST.csv
• Udacity_MAILOUT_052018_TRAIN.csv
• DIAS Attributes - Values 2017.xlsx

These files should be obtained from the Udacity workspace for this project.

Results

Results are discussed in more detail in the accompanying blog post, but in summary:

The unsupervised model identifies clusters of the population that are more likely to be customers: financially-aware people with average income, but high expenditure (high-end cars, home ownership). The columns in the dataset that relate to this are also identified.

A supervised model is built using a Logistic Regression classifier (which is found to outperform other models). It is found that model performance can be improved by using only a subset of columns: those that were found by the unsupervised model to correlate with being a customer. The model is further optimised via hyperparameter tuning.

Finally, predictions are made on a final dataset, and the results submitted to the Kaggle competition. A score of 0.622 is achieved; this is not excellent, but is better than would be achieved with a coin toss (score of 0.5)! Possible improvements are discussed in the blog post.

Licensing and Acknowledgements

Thank you to Arvato for the dataset, and to Udacity for the excellent Data Scientist Nanodegree programme.

This software is provided under the MIT licence.