Due: Friday, April 17
Consider the data collected by a hypothetical video store for 50 regular customers.
This data consists of a table which, for each customer, records the following
attributes: Gender, Income, Age, Rentals (total number of video rentals in
the past year), Avg. per visit (average number of video rentals per visit during the past
year), Incidentals (whether the customer tends to buy incidental items such as
refreshments when renting a video), and Genre (the customer's preferred movie genre). This data is
available as an Excel spreadsheet.
- Explore the general characteristics of the data, by computing the means
and standard deviations of the numerical attributes, as well as the the
distributions of male and female customers, the preferred movie genres, etc.
- Perform the the following data preparation steps on the
data (for each add a new column to the original table for comparison
- Use smoothing by bin means to smooth the values of the Age
attribute. Use a bin depth of 4.
- Use min-max normalization to transform the values of the Income attribute
onto the range [0.0-1.0].
- Use z-score normalization to standardize the values of the Rentals attribute.
- Discretize the (original, non-normalized) Income attribute based on the following categories: High = 60K+;
Mid = 25K-59K; Low = less than $25K.
- Convert the original table (not the results of part 2) into the
standard spreadsheet format. Note that this requires
converting each categorical attribute into multiple attributes (one for each
values of the categorical attribute) and assigning binary values
corresponding to the presence or not presence of the attribute value in the
original record). For example, the Gender attribute will be transformed into
two attributes, "Genre=M" and "Genre=F". The numerical attributes will
remain unchanged. This process should result in a new table with 12
attributes (one for Customer ID, two for Gender, one for each of Income,
Age, Rentals, Avg. Per Visit, two for Incidentals, and three for Genre).
- Using the standardized data set (from part 3), perform basic
analysis among the attributes. Discuss your results by indicating any
strong correlations (positive or negative) among pairs of attributes. You
need to construct a complete Correlation Matrix. Be sure to first remove the
Customer ID column before creating the correlation matrix.
- Perform a cross-tabulation of the two "gender" variables
versus the three "genre" variables. Show this as a 2 x 3 table with entries representing the
total counts. Then, use a graph or chart that provides the best visualization of the relationships
between these sets of variables. [See Slide 24 in
Understanding Characteristics of
Data for an
example. Also review Chapter 4 of Berry and Linoff.] Can you draw any significant conclusions?
- Select all "good" customers with a high value for the Rentals attribute ( a
"good customer is defined as one with a Rentals value of greater than
or equal to 30). Then, create a summary (e.g., using means, medians, and/or other statistics) of
the selected data with respect to all other attributes. Can you observe any significant
patterns that characterize this segment of customers? Explain.
Note: to know whether your observed patterns in the target group are
significant, you need to compare them with the general population using
the same metrics.
- Suppose that because of the high profit margin, the store
would like to increase the sales of incidentals. Based on your observations in
previous parts discuss how this could be accomplished (e.g., should customers
with specific characteristics be targeted? Should certain types of movies be
preferred? Etc.). Explain your answer based on your analysis of the data.
WEKA to perform the following tasks on the original data set
(use the Comma Separated version of the above data set:
Video_Store.csv). Load the data into WEKA
Explorer (the Preprocessing module). Remove the Customer ID
attribute. Review basic statistics for different attributes by clicking on the
name of each one in "attribute" panel. Next, use the unsupervised
attribute "Discretize" filter to discretize the Age attribute.
Finally, use the unsupervised attribute "Normalize" filter to
convert all of the remaining numerical attribute into [0,1] scale. Save the
resulting data set into an ARFF formatted file and submit with your answers for
the above questions.