Predictive Analytics and Modeling in Product Pricing (Personal and Commercial Insurance)

About life insurance, what it is, it's history, applications, why it is useful and more. What are the benefits for insurance companies and customers.

Brief history of predictive analytics in insurance and which risk factors are used by companies today to calculate the premiums. Example of online insurance and how they calculate it.

What are the new opportunities which arived with advancements in this scientific field recently. This is especially connected with the development of big data - we can use data in many more ways and here we discuss which new fields have opened out with those new developments.

Underwritting has been the heart of the insurance business. Here we discuss its future and how this is changing with predictive modeling.

Marketing is a big area where predictive analytics is already useful. Here we link marketing and marketing in life insurance as a special usecase for predictive analytics and modeling. We briefly describe the usecases.

Insurtech startups, possibilities of utilizing big data for insurance.

How data is organized in storing and developing the data - as well as utilizing it.

Differences and similarities between the two sciences in how they work and what they do.

The algorithms usually used in modeling and general guidance what to do once given a dataset. How the process works, what to focus on, etc.

What are the data operations that modeler does once receiving the data.

Types of data: structured and unstructured and how to recognize them - also how does certain type of data affect the modeling process.

Types of data: static and streamed and how to recognize them - also how does certain type of data affect the modeling process.

Types of variables we work with and the challenges that come with different kinds of variables.

Once we import the data, we must make sure that we can do models out of it; finding erroneus records, checking for different data issues and dealing with them.

Fixing the dimensionality of the data; reducing the amount of variables and good practices.

All about dealing with outliers and extreme cases in data.

Smoothing the data means creating a function that attemts to capture important patterns in the data.

Process of consturcting a curve or mathematical function that has the best fit to a series of data points.

This is an approach to analyzing data sets to summarize their main characteristics, often with visual methods. We explore it in this lecture.

A useful approach when modeling is the creation of dummies. Here we learn when to do this and how.

Ways how we can tranform the data.

We can split data in several ways - this is mostly done to assure proper testing of the model. Splitting is described here with its good practices - besides train and test set, we also talk about validation set and what criteria to use for splitting.

About data sources and also a brief introduction to Apache Hadoop.

About forecasting. What it is, how we can forecast. What are the goals of forecasting?

We want to know how good our prediction is. This is why we have measures of aggregate errors. We examine several most widely used measures and why is any measure useful i.e. when to use it.

What is seasonality, few ways to tacle it, how to detect it.

Measuring seasonality, some examples of tackling seasonality and about time series modeling.

Autoregressive integrated moving average method for time series data understanding and predicting (forecasting). There are several ARIMA models and the way how they work is described here.

Autoregressive conditional heteroskedasticity model and some other time series handling and forecasting models.

Linear regression - the basic forecasting model. What it is and the way how it is used.

Maths behind simple liner regression using a single predictor - how we estimate it, assesing errors, confidence intervals of predictors, etc.

Stepwise linear regression - what that is, why it's useful and exact steps of how it's done.

What it is where it's useful, how it is done.

Multivariate linear regression.

Maximum likelihood estimation is a very popular method in many models, especially in insurance. Here we go through the basic idea and how it is done in practice. This is especially useful in generalized linear models.

As an example of probabilistic model we present logit model.

As another example of probabilistic model we present the probit model. Commands in Python and R for logit and probit are also presented here.

An introduction to decision trees and how they work. The whole idea behind them is discussed here.

Here we will have some examples in scripts of how random forest algorithm works. We also discuss some more advanced topics in random forest and decision trees - that is how we learn from data, information gain for tree models, gini impurity and stopping criteria - when to stop growing a tree/forest.

Decision Tree Types, advantages and disadvantages of decision trees and about other possible algorithms in modeling.

In this exercise you will need scripting language to implement some of the models learned in this module.

Solutions given in R and Python. Together with narration.

In this exercise you will need scripting language to implement some of the models learned in this module.

Solutions given in R and Python. Together with narration.

About mortality and life expectancy modeling. We will also go through standard datasets that we can get - with some examples.

Implementation and maths behind Lee-Carter model.

Implementation and maths behind Poisson log-bilinear model.

Implementation and maths behind M7 model. This is one of the models that has been proposed more recently and is gaining traction.

Advantages & Disadvantages of different models.