Predictive Analytics 4.0 Virtual Event

Credits: 1.20 SOA CPD

Compentency: Results-Oriented Solutions

Moderator: Toby L. Hall, FSA, MAAA

Presenter(s): Jonathan Culbert

Session Description:
Artificial intelligence built on machine learning is being rapidly adopted throughout the insurance industry. From fraud detection to product design to claims processing, AI is transforming the way insurance companies do business. While the benefits of AI are significant, its acceptance also comes with risks and ethical questions. Unintentional algorithmic bias and lack of transparency can lead to unfair outcomes and reputational damage to companies. What pitfalls should practitioners of machine learning and AI seek to avoid, and what tools are available to ensure that AI is implemented as fairly and ethically as possible?

Experience Level: Beginner/Implementer

Credits: 1.20 SOA CPD

Compentency: Results-Oriented Solutions

Presenter(s): Kilian Blum; Boyi Xie

Session Description:
At some point in any predictive analytics project, managers must decide whether model performance is sufficient for deployment. Mistakes can lead to significant adverse consequences when machine learning based costing, underwriting or persistency models do not perform as expected. Furthermore, under certain circumstances, metrics like accuracy can be deeply misleading when assessing model performance. This session will not only investigate the different metrics but also look at case studies to illustrate how metrics behave under unique circumstances. Most importantly, these case studies will show how the underlying business objective determines which metric should be focused on.

Experience Level: Manager/Supervisor

Credits: 1.20 SOA CPD

Compentency: Technical Skills & Analytical Problem Solving

Moderator(s): Rosmery Cruz

Presenter(s): Guizhou Hu

Session Description:
Age and face amount conventionally drive underwriting decisions, but what happens when the underwriter encounters two applicants of differing medical profiles? Should the underwriter be required to pull a complete set of medical information for both? In this presentation, we will introduce a more dynamic approach to assessing underwriting requirements by integrating predictive models with actuarial protective value assessments. This new approach streamlines accelerated underwriting programs, saving time and money for the carrier and applicant alike. Find out how insurers can transition from conventional AA requirement concepts into a more individualized underwriting paradigm.

Experience Level: All

Credits: 1.20 SOA CPD

Compentency: Technical Skills & Analytical Problem Solving

Moderator(s): Jason Rodriguez

Presenter(s): Dan Kim, FSA, CERA, MAAA

Session Description:
Predictions will likely be inaccurate as a model is a generalized version of the reality. In this session, the presenter will discuss a framework to understand and quantify prediction uncertainties. They will discuss statistical methods such as Frequentist and Bayesian approaches to quantify model uncertainties. Lastly, the presenter will share some practical considerations such as experimental designs, interpretation of models and outputs to inform the level of prediction uncertainty.

Experience Level: Manager/Supervisor

Credits: 1.20 SOA CPD

Compentency: Technical Skills & Analytical Problem Solving

Presenter(s): Shea Parkes, FSA, MAAA

Session Description:
Gradient Boosting Machines (GBM) are effective with many problems. They strike a great balance between accuracy and interpretability. This session will be an ode to the GBM. It will include a bit of history about who came up with the original idea, who helped refine the idea into statistical learning, and who helped write the amazing modern-day implementations. We'll ask the attendees to share their own love stories or hate mail if they're brave enough.

Experience Level: Beginner/Implementer

Credits: 1.20 SOA CPD

Competency: Technical Skills & Analytical Problem Solving

Presenters: John Myslinski, ASA

Session Description:
There are several statistical and machine learning techniques for modeling mortality. In this session, we will discuss the various approaches and demonstrate how you can use the techniques from Generalized Additive Models (GAM) to model mortality. The session will be structured as a case study of a pricing actuary setting the mortality assumption for a product. The presenter will walk through the framework for building a GAM in R with the mgcv-package, highlighting major modelling decision points along the way. We will then wrap up the session with an overview of ways to validate your resulting mortality assumption.

Experience Level: Advanced Practitioner

Moderator(s): Xiaojie Wang, FSA, CERA

Session Description:
Accelerated Underwriting remains the most common application for predictive analytics within insurance companies. This moderator led discussion will inform you of what tools and methodologies have been successful, common challenges and pitfalls to creating and maintaining an accelerated underwriting program. Regulatory issues will also be discussed as the industry weighs the benefits of making insurance more accessible against having less transparency from model-based decisions.

Moderator(s): David Moore, FSA, MAAA

Session Description:
Insurance uses several data points on an applicant to assess their risk, and pools them with other similar risks to manage the risk and provide a fair price to the insured. The advances in data collection and advanced analytics enable us to understand risk profiles better than ever before, but as we automate the pricing and underwriting process, are we using big data ethically to do this, and are we inadvertently using biased models to make underwriting decisions? This session will be an open discussion around what we as actuaries can and should do when using predictive models in an ethical manner.

Credits: 1.20 SOA CPD

Compentency: Results-Oriented Solutions

Presenter(s): Jeff T. Heaton, Ph.D.

Session Description:
Machine learning experts are attempting to use automation to solve labor shortages across many industries. Within the insurance industry, tasks such as underwriting, fraud detection, forecasting, and marketing could benefit from increased efficiency through both machine learning automation and augmentation. Machine learning experts also apply this same automation to more routine job tasks through a process known as Automated Machine Learning (AutoML). With AutoML you can augment a data science team or capture data even if you do not have access to data scientists. This trend has many positive benefits including data democratization and the rise of citizen data scientists. Moreover, this presentation will provide an overview of both commercial and open-source AutoML and the platforms strengths and weaknesses.

Experience Level: Manager/Supervisor

Credits: 1.20 SOA CPD

Compentency: Technical Skills and Analytical Problem-Solving

Presenter(s): Michael Cletus Niemerg, FSA, MAAA

Session Description:
In this session, explore the application of predictive models in pricing stop-loss coverage. The presenter will first explain the differences between stop-loss coverage and fully-insured coverage, why predictive models have the potential to increase pricing accuracy, and why the nature of stop-loss coverage makes prediction particularly difficult. Followed by considerations and challenges in adapting fully insured cost models for use in stop-loss, as well as techniques and considerations when building models customized for stop-loss. Lastly, we will cover general pricing considerations, hurdles to adaptation, and practical considerations in the development and use of predictive models.

Experience Level: Advanced Practitioner

Credits: 1.20 SOA CPD

Compentency: Technical Skills and Analytical Problem-Solving

Presenter(s): Martha Grace; Mark Andrew Sayre FSA, CERA

Session Description:
As predictive models become more embedded within the industry, regulators have become increasingly concerned about the use of these models in areas such as underwriting, pricing, and claims adjudication. Actuaries must play a critical role in ensuring that companies are able to defend the ways in which models are being used. This session will provide attendees with a proposed framework for actuarial justification that can be applied when considering advanced predictive models. The discussion will begin with simpler model designs including logistic regression or GAMs and will proceed into more advanced designs with random forests. Techniques for assessing fairness and measuring potential adverse impact will also be outlined.

Experience Level: Advanced Practitioner

Credits: 1.20 SOA CPD

Compentency: Relationship Management and Interpersonal Collaboration

Presenter(s): Jordan Durlester; Bradley Joseph Lipic

Session Description:
Insurers across the globe are becoming obsessed with identifying the best opportunities to use data strategically across the value chain. At the same time, new legislation is moving from the discussion phase to the implementation phase, raising compliance issues to be addressed. Data and analytics leaders are faced with the dual challenges of optimizing data awareness and use-case identification (playing offense) while balancing reputation and regulatory risk (playing defense). In this session, Brad Lipic and Jordan Durlester will walk you through key considerations in designing and implementing a data marketplace that allows your organization to optimize value while properly managing risk.

Experience Level: Manager/Supervisor

Credits: 1.20 SOA CPD

Compentency: Professional Values

Presenter(s): Kevin J. Pledge, FSA, FIA; Neil Raden

Session Description:
Context is critical when it comes to AI ethics. This panel of experts will present current issues relating to ethical use of AI in insurance, followed by interactive case studies.

Experience Level: Beginner/Implementer

Credits: 1.20 SOA CPD

Competency: Technical Skills and Analytical Problem-Solving

Moderator: Patrick Jason Colbert, FSA, MAAA

Presenter(s): Lindsay Allen, ASA

Session Description:
Geographic Information Systems (GIS) can be used for problem solving and decision making. Geospatial analyses can be a helpful way to identify relationships, spot trends and make inferences that may not be apparent in tabular form. This presentation will overview how to use open source packages in R or Python for mapping analyses, such as interactive choropleth maps and drive times. Also, throughout the presentation the CMS Data Entrepreneurs' Synthetic Public Use File will be used to illustrate concepts where appropriate.

Experience Level: All

Credits: 1.20 SOA CPD

Competency: Results-Oriented Solutions

Presenter(s): Clinton Innes, FSA, ACIA

Session Description:
Data science applications to life insurance and disability insurance claims is a growing area of interest. This session will give a deep dive into some of the applications and techniques that are relevant to the emerging field of claims analytics. It will cover topics including fraud analytics, approval and denial models, and claim scoring for disability insurance. Additionally, this session will highlight how actuaries and data scientists can collectively leverage analytics to improve both customer and claims experience.

Experience Level: All

Credits: 1.20 SOA CPD

Competency: Technical Skills and Analytical Problem-Solving

Presenter(s): Robert Jason Reed, FSA, MAAA

Session Description:
Many advanced modeling techniques from data science are filtering into actuarial practice including claims forecasting, trend analysis and risk assessment. The number of models continues to grow, but frequently the choice of which to use is arbitrary. There are few good theoretical reasons to prefer one over the other. Ensemble learning is an emerging field within data science that combines models to form more stable and accurate predictions, often through voting on the best one. However, the increased accuracy of these models comes with more difficulty in interpreting output and understanding true drivers of predictive performance. Learn how ensemble machine learning methods are being applied to make better forecasts.

Experience Level: Manager/Supervisor

Moderator(s): Michael Cletus Niemerg, FSA, MAAA

Session Description:
This will be a moderator led discussion is on how actuaries and actuarial students new to predictive analytics can get started with the tools and applications needed to make an impact in this emerging and rapidly evolving area. Questions that will drive the discussion are, how can you generate support and funding to develop new and innovative tools? How can you find a mentor if no one at your organization is using Predictive Analytics tools? Practitioners of all experience levels and invited to attend and share their thoughts on best practices for learning and development.

Moderator(s): David Moore, FSA, MAAA

Session Description:
Insurance uses several data points on an applicant to assess their risk, and pools them with other similar risks to manage the risk and provide a fair price to the insured. The advances in data collection and advanced analytics enable us to understand risk profiles better than ever before, but as we automate the pricing and underwriting process, are we using big data ethically to do this, and are we inadvertently using biased models to make underwriting decisions? This session will be an open discussion around what we as actuaries can and should do when using predictive models in an ethical manner.

Credits: 1.20 SOA CPD

Compentency: Technical Skills and Analytical Problem-Solving

Presenter(s): Matthew Kevin Heaphy, FSA, MAAA

Session Description:
Any actuary using R should familiarize themselves with R Markdown. R Markdown allows users to create beautifully-designed reports that weave in both text and code to generate tables, plots and other elements using a very simple syntax. These documents can take a variety of formats including memos, presentations and dashboards. Best of all, R Markdown documents are fully reproducible, transparent and auditable. Join us for this session if you want to take your reports to the next level.

Experience Level: All

Credits: 1.20 SOA CPD

Compentency: Technical Skills and Analytical Problem-Solving

Presenter(s): Dihui Lai, ASA

Session Description:
Natural language processing (NLP) has been widely used in various applications, including email, search engines and mobile applications. Interest is now growing in technology applications for insurance, and yet some question whether NLP technologies today are mature enough. The presentation will discuss the latest NLP advances, the limitations of NLP and lessons learned in applying NLP to life insurance use cases.

Experience Level: Advanced Practitioner

Credits: 4.8 SOA CPD/ 4.0 CIA

For those looking for some truly hands-on experience, or those looking to demonstrate their machine learning expertise through a friendly competition, the Predictive Analytics and Futurism Section is proud to once again host the half-day Hack-A-Thon. Attendees who choose to participate in the Hack-A-Thon will compete on teams to build the most accurate model of a real dataset. The Hack-A-Thon is open to attendees of all skill levels, but this year we have teamed up with the PPAS to make this a great learning opportunity for beginners as well. This event will be fast, furious and fun, with four hours to develop a model and submit predictions.

Contest participants must be registered for Predictive Analytics 4.0.

Agenda

• 10:30 – 11:00 AM CT: Pre-Hack-A-Thon session (discussion of rules, download of data, answer questions from participants, etc.)
• 11:00 AM – 3:00 PM CT: Hack-A-Thon Contest
• 3:00 – 3:30 PM CT: Announcement of winners and recap

Official Content Rules