Explore BrainMass

Clustering and Market Basket analysis

Retro Motors is considering offering predefined options combination packages (trim levels) for its cars rather than letting customers select individual options. The company wants to determine what the most desirable options packages are. Sven Jorgensen, Retro data analyst, has been looking at options packages and working with clustering as a data mining technique.
Market Basket Analysis (MBA) would be appropriate for the problem and agrees to send you an e-mail containing a brief analysis of the problem.

From Sven Jorgensen
Subject: Options Packages Data Mining Report
As you know, Retro is interested in offering different trim packages for its models (similar to Honda's DX, LX, and EX model trim levels). Currently, customers are allowed to choose individual options and accessories when they place a custom order. Enrique Munoz, the manufacturing manager, thinks that more standardized trim levels will increase efficiencies in his department and provide better market segmentation for sales. We need to figure out what these different trim levels should be. I realize that the trim levels must satisfy most customers' preferences while still being profitable for our company.
I have completed an initial analysis of our sales data. Based on our discussions, I thought this situation would be a great opportunity to apply some data mining techniques. In fact, it appeared to me that the data are well suited for market basket analysis, so I applied that technique. Here are the association rules with the highest support:
Rule Confidence Support
Turbo, V-8, Bucket seats Leather-wrapped steering wheel
1.0 0.056
Rain detector, Fog lamps Parking sensor
0.8 0.056
Leather seats, Heated seats Power seats
0.8 0.056
White letter tires 1.0 0.052
Parking sensor 1.0 0.047
Pinstriping, White letter tires Custom wheels
0.76 0.047
Power seats Heated seats
0.60 0.042
Heated seats Heated mirrors
0.22 0.039
Custom wheels White letter tires
0.25 0.037
Power locks Bucket seats
0.62 0.037
After finishing this analysis, I am more confused than ever. I still cannot figure out which options should apply to the proposed new trim levels. Do you have any ideas?

From: Sean Canada
Subject: Options Packages
Thanks for contacting me about your data mining project regarding the trim styles. Here are my thoughts on this issue.
Yes, Sven's list may show what options are selling together, but it would leave many Retro customers unsatisfied, especially for the Vortex. We sell leather seats on 48 percent of the Vortexes, not standard upholstery. Vortex customers spend between $15,000 and $23,000, depending on the options they add. The low-end Vortex has a very different list of options than the high-end Vortex. And you can probably find a middle group, too. Why don't you look at what the options packages would be at three different price points for each model? If you look at it that way, it would really help Retro use the three different trim levels to target different market segments more effectively.

Alisa Quinn, Appex Decision Support
Subject: Market Basket Analysis

In our recent conversation, you indicated that Retro is contemplating introducing three trim levels for its new models. I started with the market basket analysis provided by Mr. Jorgensen, and took into consideration the customer information from your past sales. I performed an initial clustering of the data to see whether any obvious patterns emerged. Then I performed a market basket analysis on the clustered data.
The results, as you will see from the attached report, were quite interesting. I think that the results from this combination analysis will be more helpful to you in determining options packages, because of the more targeted nature of the data mining process I used. Let me know if you have any questions about the analysis.
Thank you for the opportunity to provide data mining services to Retro Motors. If I can be of further assistance, please feel free to contact me.
Alissa Quinn

Options Packages Report
In response to your request for data mining assistance related to options packages for Retro Motors, I first clustered sales by customer age and salary, and found three distinct clusters in the data. For simplicity's sake, I labeled the clusters "Entry," "Sport," and "Luxury." Then I performed a market basket analysis to discover the popular options combinations within each segment. The figure below is a graphic representation of the clustering process.

Market Basket Analysis
The following automotive options were considered for the market basket analysis:
Power windows Power door locks
Turbo Rust-proofing
Leather seats Floor mats
Rain detector V-6 engine
Parking sensor V-8 engine
Fog lamps Heated mirrors
Pin striping Heated seats
White letter tires CD player
Custom wheels Six-CD changer
Halogen lights Heads-up display
Leather-wrapped steering wheel Steering wheel controls
Power seats
Cluster: Luxury
The following association rules were generated for the Luxury trim level:
Association Rule Confidence Support
Leather seats, Heated seats Power seats
1.0 0.113
Rain detector, Fog lamps Parking sensor
0.91 0.113
Power seats Heated seats
0.89 0.09
Heated seats Heated mirrors
0.88 0.085
Heated mirrors Heated seats
0.6 0.067
Cluster: Sport
The following association rules were generated for the Sport trim level:
Association Rule Confidence Support
Turbo, V-8, Bucket seats Leather-wrapped steering wheel
1.0 0.115
Pin striping, White letter tires Custom wheels
1.0 0.092
Power locks Bucket seats
0.88 0.08
Custom wheels White letter tires
0.86 0.069
Parking sensor Fog lamps
0.99 0.068
Cluster: Entry
No interesting association rules were discovered

See attached for full problem description.


Solution Preview

Here the automobile company wants to identify the characteristics of compact car that potential purchasers considers it very important.

Clustering is a technique useful for exploring data. It is particularly useful where there are many cases and no obvious natural groupings. Here, clustering data mining algorithms can be used to find whatever natural groupings may exist.

Clustering analysis identifies clusters embedded in the data. A cluster is a collection of data objects that are similar in some sense to one another. A good clustering method produces high-quality clusters to ensure that the inter-cluster ...

Solution Summary

This explains the clustering and market basket analysis with the help of the case of Retro Motors