Description: A look into how machine learning models implement classification and regression at a deeper level. Covers topics such as multi-regression, softmax, and ROC charts.

Description: The newer predictive modeling tools are powerful but they work best when supported by a lot of computing power. Parallel processing is very powerful but often very difficult to implement. Dihui and Richard take us through the maze of options with H2O, SparkNet/MLib and Mahout. Which do you use for a generalized linear model (GLM), and which is better for high-volume data manipulation?

Description: This actually is a copy of an appendix of a research paper sponsored by the SOA Committee on Finance Research. You can read the entire research paper: “Lapse Modeling for the Post-Level Period—a Practical Application of Predictive Modeling” at the SOA site https://www.soa.org/Research/Research-Projects/Finance-Investment/lapse-2015-modeling-post-level/#sthash.W9lERSls.dpbs, but you can read this valuable appendix copied here in this issue to see how to build a PM step-by-step using your data and the R programming language. We thank Richard Xu, Dihui Lai, Minyu Cao, Scott Rushing, and Tim Rozar for their excellent paper and the Society of Actuaries, for the permission to reprint this portion of the paper.

Description: Description of the vast variety of predictive analytics topics covered by the Forecasting & Futurism section and introduction to the July 2015 issue of the newsletter.

Description: Each month in the media we learn of another set of jobs being automated into oblivion or replaced by disruptive technologies. We see taxi drivers replaced by Uber and Lyft now, and soon by self-driving cars. An Oxford University study of 702 occupations showed insurance underwriters in 698th place, with a 99% probability of computerization. They were safer for obsolescence by automation than only four occupations such as hand sewers and telemarketers.

Description: In this article, we will overview some popular tree-based models and understand how such models may be applied to the insurance industry.

Description: In An Introduction to Incremental Learning, by Qiang Wu and I describe how machine learning is a natural enhancement to predictive analytics through its many useful tools to derive meaning from a lot of otherwise unusable data such as handwriting. Again, there are formulas involved; but we have tried to keep them understandable to the actuary who isn’t already familiar with stochastic gradient descent, perceptrons, and principal component analysis. Incremental learning can help you refine the estimations you derive from more traditional actuarial methods.

Description: The article introduces the concept of overfitting in predictive modeling and illustrates how it hinders our ability to predict new data well.

Description: This article discusses techniques used to create risk scores to help determine appropriate primary care panel sizes for a provider organization.

Description: The editor of the newsletter provides a running list of articles included in the issue highlighting such topics as the applicability genetic algorithms, the power of group dynamics, judgmental forecasting, improvement in mortality and complexity science.