Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA, and Michael Niemerg, FSA, MAAA, about artificial neural networks. Artificial neural networks have gone through a recent renaissance as part of deep learning. Artificial neural networks can now provide amazing accuracy, but they come with an equally amazing amount of complexity. This podcast provides a very high-level overview of everything from their history to modern implementations.

Direct download: SOA17_29_PAF_08__Neural_Nets_Renaissance.mp3
Category:Predictive Analytics and Futurism -- posted at: 11:42am CDT

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about Gradient Boosting Machines (GBMs). Although GBMs share many characteristics with Random Forests, they can often provide better performance at the cost of somewhat higher complexity. This discussion covers topics including underlying theory, tuning hyperparameters and modern implementations.

Direct download: SOA17_19_PAF_07_GBMs.mp3
Category:Predictive Analytics and Futurism -- posted at: 10:42am CDT

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about how to do well at Kaggle. In the last episode they explored what Kaggle is and what makes up the various predictive modeling contests they host. In this episode, Shea shares some tips and general advice on how to perform well in a given Kaggle contest.

Direct download: SOA17_17_PAF_05_How_to_do_Well_at_Kaggle.mp3
Category:Predictive Analytics and Futurism -- posted at: 3:42pm CDT

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about Kaggle. Kaggle is an online community that brings together data scientists from across the globe to participate in predictive modeling contests.

Direct download: SOA17_17_PAF_05_Kaggle_Competition.mp3
Category:Predictive Analytics and Futurism -- posted at: 3:38pm CDT

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about Random Forests.  Random Forests are a very important machine learning algorithm that all practitioners should be very comfortable with. They provide great performance with minimal tuning and headache. As a bonus, this session features what is hopefully much improved audio.

Direct download: SOA_17_02_PAF_02_-_Random_Forests.mp3
Category:Predictive Analytics and Futurism -- posted at: 8:54am CDT

Listen in as Shea Parkes, FSA, MAAA interviews Timothy Paris, FSA, MAAA and Steve Wright, FSA, MAAA from Ruark Consulting.  Tim and Steve have developed predictive analytics expertise at their company over the last few years.  They share stories from that journey and some suggested resources for others wanting to do the same.


Predictive Analytics and Futurism Section: Ensemble Predictive Modeling

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about ensemble predictive modeling.  The most accurate predictive models are not any single predictive model; they are ensembles of different models.  This discussion explores what types of ensembles are common, and some of the theory that explains their surprising power.  Understanding some of this theory can help you identify problems or individual models that might benefit most from ensembling.


Predictive Analytics and Futurism Section: Decision Trees

Listen in as Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA about decision trees.  Decision trees are an important class of predictive models that are not often covered in traditional statistics classes.  Although it is becoming more rare to use a single decision tree, they are an important building block in at least two of the most prevalent ensemble learning methods in use today.  This discussion tries to lay the groundwork so the next sessions can focus on RandomForests and Gradient Boosting Machines.


Predictive Analytics and Futurism Section: Penalized Regression

Anders Larson, FSA, MAAA, interviews Shea Parkes, FSA, MAAA, and Michael Niemerg, FSA, MAAA, about penalized regression.  Penalized regression can be thought of as the result of applying actuarial credibility to linear modeling.  They explain the statistical theory behind it and the practical reasons an actuary might apply it.  They also touch on some useful implementations that are freely available as open source libraries for R and Python.


Predictive Analytics and Futurism Section: Predictive Modeling Case Study – Life Insurance Underwriting

Listen in as Shea Parkes, FSA, MAAA, from Milliman interviews Vince Granieri, FSA, MAAA, EA from Predictive Resources.  Vince presents a case study of applying predictive analytics to a classical actuarial problem: life insurance underwriting.  Vince walks us through building up his team, his capabilities and his models to provide more useful solutions with predictive analytics.