Description
In this project we will compare the results of using clustering algorithms to segment customers
based on their purchasing data.
The project may be completed individually, or in a group of no more than three (3) people. All
students on the team must be enrolled in the same section of the course.
Platforms
The platform requirements for this project are the same as for previous projects.
Libraries
You will need pandas to load the dataset, scikit-learn to run clustering algorithms, and SciPy to
visualize hierarchical clustering with a dendrogram.
You may reuse code from the Jupyter notebooks accompanying the textbook and from the
documentation for the libraries. All other code and the results of experiments should be your
own.
Datasets
Download groceries.csv, a dataset consisting of weekly spending by families on several
common grocery items. Spending is listed in units of dollars per person per week.
Experiments
Run the following experiments in a Jupyter notebook, performing each action in a code cell and
answering each question in a Markdown cell.
1. Use read_csv() to load and examine the dataset.
2. Use sklearn.cluster.KMeans to cluster the dataset using the default parameters.
Assign the labels_ attribute determined by the algorithm to a new column in your
DataFrame. How many clusters are there?
3. Use pandas.DataFrame.groupby() to group the data by cluster assignment, then use
the GroupBy object to examine descriptive statistics such as minimum, maximum, and
mean. Describe any differences you see between the clusters.
4. There’s no particular reason to think that the default value of K is the correct one for this
dataset. Let’s switch to hierarchical clustering to see if we can visualize how the data
clusters together.
Plot a scipy.cluster.hierarchy.dendrogram() for the dataset. Single linkage may
not be the best choice for the dataset, but it is relatively quick to compute and will help
us make a decision on the number of clusters.
Based on the dendrogram, how many clusters appear to be present in the dataset?
5. Repeat experiment (2) using the number of clusters you determined to be present in
experiment (4).
6. Now switch to sklearn.cluster.AgglomerativeClustering() and repeat for the
same number of clusters. How do the label assignments compare for the two
algorithms?
Caution: don’t include the cluster number assigned by K-Means as one of the features to
be clustered, or it will skew your results.
7. Repeat experiment (3) with the clusters you obtained in experiment (5). How would you
describe the various types of customers? What can you determine about them based on
the data? For each segment identified, list some other items they are and are not likely
to buy in future shopping trips.
Submission
Submit your Jupyter .ipynb notebook file through Canvas before class on the due date. Your
notebook should include the usual identifying information found in a README.TXT file.
If the assignment is completed by a team, only one submission is required. Be certain to identify
the names of all students on your team at the top of the notebook.
