2017 Intermediate Predictive Modeling Limited Attendance Seminar

Registration for this event is now closed.

Registration for the Introductory Predictive Modeling LAS must be done separately.

Generalized Linear Models (GLMs) and such extensions as Generalized Additive Models and Multilevel/Hierarchical Models are today considered standard elements of the actuarial toolkit. As the size and complexity of available datasets continue to grow, actuaries and data scientists increasingly find it useful to adopt a broader array of statistical and machine learning tools. The Intermediate Predictive Modeling Limited Attendance Seminar [PMLAS-2] is intended for actuaries and data scientists who wish to complement a working knowledge of GLMs with other tools.

The techniques covered in this seminar fall into three classes. First, a variety of methods are available for achieving parameter shrinkage ("credibility weighting") within GLMs. This seminar will cover two highly practical and increasingly popular such methods: ridge and lasso regression. Second, machine learning techniques offer a powerful way of discovering and modeling interactions and nonlinear relationships in complex datasets. The seminar will cover some of the more popular machine learning techniques. Finally, the seminar will cover a variety of unsupervised learning methods (learning methods that do not involve a pre-specified outcome variable). Such methods are useful for identifying clusters and reduced-dimension data features in complex datasets. A common theme of all three classes of techniques is harnessing computing power to make sense of "wide" datasets, which commonly contain hundreds or even thousands of columns.

The seminar is designed to provide a general survey of techniques which complement the GLM framework. A substantive case study, as well as a number of smaller case studies, will complement the theoretical discussion with hands-on experience applying the techniques.

Learning objectives:

The seminar will help participants

• Gain an understanding of the strengths, weaknesses and typical uses of various statistical and machine learning techniques
• Gain a working knowledge of how to apply the predictive modeling techniques through the use of examples and case studies with real-world significance
• Gain an understanding of the inputs and outputs of the techniques and how to gauge the models’ effectiveness.

Note that this seminar presupposes working knowledge of Generalized Linear Models, as well as comfort manipulating data and fitting models in the R statistical computing environment.

Seminar content:

• Shrinkage methods: Regularized regression (ridge and Lasso regression). Conceptual discussions of Bayesian regression and multilevel/hierarchical models will be provided to help contextualize the techniques.
• Tree-based Models: Classification and Regression Trees [CART], Random Forests, Gradient Boosted Trees
• Other supervised learning methods: Support Vector Machines, Multivariate Adaptive Regression Splines, ensemble methods
• Topics in unsupervised learning: k-means and hierarchical clustering, mixtures of Gaussians, principal components analysis. Other techniques such as matrix completion, association rules, and multidimensional scaling will be touched on time permitting.

Attendance is limited to a maximum of 35 students. Attendees will be selected on a first-registered, first-accepted basis. Participants are expected to bring their own laptop, loaded with the R software, to the seminar.

CONTINUING EDUCATION CREDIT

The CAS Continuing Education Policy applies to all ACAS and FCAS members who provide actuarial services. Actuarial services are defined in the CAS Code of Professional Conduct as "professional services provided to a Principal by an individual acting in the capacity of an actuary. Such services include the rendering of advice, recommendations, findings, or opinions based upon actuarial considerations." Members who are or could be subject to the continuing education requirements of a national actuarial organization can meet the requirements of the CAS Continuing Education Policy by satisfying the continuing education requirements established by a national actuarial organization recognized by the Policy.

Participants should claim credit commensurate with the extent of their participation in the activity. CAS members earn 1 CE Credit per 50 minutes of educational session time, not to include breaks or lunch.

Note: The amount of CE credit that can be earned for participating in this activity must be assessed by the individual attendee. It also may be different for individuals who are subject to the requirements of organizations other than the American Academy of Actuaries.

CANCELLATIONS

Registrations fees will be refunded for cancellations received in writing at the CAS Office via fax, 703-276-3108, or email, refund@casact.org, by November 22, 2017 less a $200 processing fee.

CONTACT INFORMATION

• For more information on content, please contact Jody Allen, Professional Education and Research Coordinator, at jallen@casact.org.
• For more information on attendee registration, please email the Actuaries' Resource Center at arc@casact.org.
• For more information on the Seminar other than registration or content issues, please email meetings@casact.org.
• For more information on other CAS opportunities or regarding administrative policies such as complaints and refunds, please contact the CAS Office at (703) 276-3100 or office@casact.org.