A Prediction Model for High-Dollar Black Friday Shoppers

Overview

Black Friday is the most significant retail shopping day in America and being able to plan strategy around attracting high-dollar buyers would be an important advantage for retailers in an increasingly competitive retail market. Using a Kaggle dataset and Alteryx, I created three predictive models for finding shoppers likely to spend more than $10,000 on their Black Friday shopping. 

Description of the Data

The dataset for this project was obtained from Kaggle and uploaded into Alteryx. I used Alteryx to examine the data, looking for missing values or outliers. None were found in the data, and the data seemed to be very clean, so little to no cleaning the data was necessary. The dataset contained 537,577 records. Given the cleanliness of the data, I proceeded to explore the data.

Model Selection

I created multiple models in Alteryx, including boosted tree, decision tree, and neural network models. The model results are as follows:

-       

           - The Boosted Tree produced  more accurate results than the decision tree model. The confusion matrix for the decision tree and boosted tree models are shown below:

Decision Tree:

Boosted Tree:

     -

I also used Alteryx to construct a variable importance plot which can be seen below:

   

We can see in the plot that gender, occupation, and city category are the most influential factors in predicting whether or not a person will be a high-dollar shopper. However, the chart does not indicate the direction of the correlation. To explore these categories more, I created plots of means. The plots can be found below: 

The plots of means indicate that buyers with jobs 12, 15, and 17, buyers from C type cities, and male buyers spend more than others. Using this information, a company seeking to maximize their appeal to high-dollar buyers would target these demographic groups.