PA3Undergraduate courses
Predictive Analytics III — Dimension Reduction, Clustering and Association Rules
Key unsupervised learning techniques of association rules – principal components analysis, and clustering – and will include an integration of supervised and unsupervised learning techniques.
- Intermediate
- 4 weeks
- Approx. 15 hours per week
- 3 semester hours
About this course
This course covers the key unsupervised learning techniques: dimension reduction with principal components analysis, cluster analysis for finding groups of similar records, and association rules for finding what goes with what in transaction data. Unsupervised methods have no outcome variable to be judged against, so the work is in deciding what a component or a cluster means and whether it is real. The final week puts the two halves together — using unsupervised methods to shrink a predictor space or to split records into homogeneous groups before a supervised model is fitted — and introduces network and text analytics, where those methods do much of the work.
Who this course is for
Marketers who need to specify customer segments and identify associations among products purchased, environmental scientists clustering observations, analysts who have to find the key variables among many, MBAs refreshing their quantitative technique, and managers and scientists who want to see what data mining can do.
What you will learn
8 outcomes
By the end of this course, you will be able to:
- Understand the issues that arise from using too many predictors — the curse of dimensionality
- Detect information overlap using domain knowledge, data summaries and charts
- Use principal components analysis to reduce predictors to fewer components of correlated predictors
- Use hierarchical clustering and k-means clustering to find and describe clusters of similar records
- Validate clusters and judge within-cluster homogeneity
- Use association rules to find patterns of what goes with what in transaction data
- Build item-based and person-based collaborative filtering recommendations
- Combine unsupervised and supervised learning methods in a final project
Week by week
4-Week curriculum overview
This curriculum is identical across all start dates. Expand a week to explore the topics covered.
- Weeks
- 4
- Your week
- Approx. 15 hours
- Level
- Intermediate
Dimension reductionWeek 1
- Detecting information overlap using domain knowledge and data summaries and charts
- Removing or combining redundant variables and categories
- Dealing with multi-category variables
- Automated dimension reduction — principal components analysis (PCA)
- Predictive algorithms with variable selection techniques
Cluster analysisWeek 2
- Popular uses of cluster analysis
- Clustering approaches
- Hierarchical clustering — distances between records, distances between clusters, dendrograms
- Validating clusters; strengths and weaknesses
- K-means clustering — initializing the k clusters, distance of a record from a cluster
- Within-cluster homogeneity and elbow charts
Association rules and recommender systemsWeek 3
- Discovering association rules in transaction databases — support, confidence and lift
- The apriori algorithm and its shortcomings
- Collaborative filtering — item-based
- Collaborative filtering — person-based
Integrating supervised and unsupervised methods; introduction to network and text analyticsWeek 4
- The role of unsupervised methods in predictive analytics — dimension reduction of the predictor space
- Predictive models on subsets of homogeneous records
- Advantages and weaknesses of combining unsupervised and supervised methods
- Network analytics
- Text analytics
- Unsupervised methods used in network and text analytics
Instructors
Expert-Led Guidance
Each cohort is led by dedicated instructors and assistant teachers who actively lead weekly discussions, provide personalized feedback and grade your assignments.
Mr. Anthony Babinec
BA in Sociology and MA in Sociology, with a focus on advanced statistics and political sociology, from the University of Chicago. He serves on the editorial board of the Journal of Targeting, Measurement and Analysis for Marketing, and has presented at the AMA's Applied Research Methods Conference, the Advanced Research Techniques Forum, the Sawtooth Software Conference and Statistical Innovation's Statistical Modeling Week.
Before you start
What you need to know first
Predictive Analytics I — Machine Learning Tools
Course Format & Schedule
This is a 4-week, 100% online, asynchronous course.
- No mandatory live sessions: Log in and complete your work at times that fit your schedule.
- Weekly Releases: At the start of each week, you will receive new lecture materials and answer keys for the previous week's exercises.
- Interactive Community: Work through exercises, submit assignments, and engage with your instructor and peers via a private discussion board.
Homework
Short-answer questions testing the concepts and guided data analysis problems using software, alongside supplemental video lectures. The end-of-course data modeling project is a capstone that combines an unsupervised task with the supervised methods of the earlier predictive analytics courses.
Texts
Please choose one of the following textbooks based on the software you plan to use:
- Python: Machine Learning for Business Analytics: Concepts, Techniques, and Applications in Python (2nd ed., 2025) by Shmueli, Bruce, Gedeck, and Patel. Also available at Amazon here.
- R: Machine Learning for Business Analytics: Concepts, Techniques, and Applications in R (2nd ed., 2023) by Shmueli, Bruce, Gedeck, Yahav, and Patel. Also available at Amazon here.
- Analytic Solver Data Mining (previously XLMiner): Machine Learning for Business Analytics: Concepts, Techniques, and Applications in Analytic Solver Data Mining (4th ed., 2023) by Shmueli, Bruce, Deokar, and Patel. Also available at Amazon here.
The same text is also used in Predictive Analytics I — Machine Learning Tools and Predictive Analytics II — Neural Nets and Regression. So one copy covers all three courses.
Software
This is a hands-on course in which you will apply data mining algorithms to real datasets. The course can be completed using Python or R, both of which are free, open-source programming languages. Corresponding editions of the course text are available for Python and R, making these the recommended options for completing the course without additional software costs. Worked examples are also available in Analytic Solver Data Mining (ASDM), an add-in for Microsoft Excel. If you choose this option, you will need both Microsoft Excel and ASDM. Course participants will receive a license for ASDM for nominal cost — this is a special version for this course. IMPORTANT: Do NOT download the free trial version available at solver.com
FAQ
Can I take this without having done the earlier predictive analytics courses?This course
The first three weeks stand on their own, since unsupervised methods need no outcome variable. Week four is where supervised and unsupervised methods are combined, and the capstone project expects you to fit a predictive model on the output of one of them, so some prior familiarity with supervised learning is recommended.
Is there a money-back or satisfaction guarantee?Every course
We handle cancellation and refund requests on an individual basis. If a course is not meeting your expectations or your circumstances change, please reach out to our team (support@learnstatistics.org) so we can work with you on a solution.
Can I transfer to a later start date or withdraw after the course begins?Every course
Yes. If you need to defer your enrollment to a future start date, withdraw mid-course, or transfer your seat to a colleague, please contact us directly. We address these requests flexibly on a case-by-case basis.
Who teaches the course?Every course
Each cohort is led by dedicated instructors and assistant teachers who actively lead weekly discussions, provide personalized feedback and grade your assignments.
Something not answered here? Ask us before you book — and the terms set out what a purchase covers.
Credit, and where it counts
This course carries a verified figure of 3 semester hours. A credit recommendation from the American Council on Education says what a course is worth; the institution receiving it decides whether to award it.
Anyone can take this subject: there is no application, and you do not have to be studying at a university. It also counts toward a degree at Thomas Edison State University, which decides what to award for it. The full position — including the limits each university publishes is on its own page, and worth reading before you buy.