Home

Starting a telematics-based car insurance company from scratch comes with a host of interesting data science challenges. The lack of historical data makes it difficult to construct accurate predictive models. On the other hand, dealing with a flood of smartphone sensor data requires us to build large-scale systems that ride the big data wave rather than being submerged by it. In this talk, we will take a whirlwind tour of topics like adverse selection, telematics risk-scoring, pricing and reserving models, and how we think about industry research projects. Time permitting, we will present a method for fitting tree-based models with shape constraints.

What happens when you raise the state minimum wage? Did your advertising campaign succeed in raising revenue? Does investing in stem cell research actually lead to more research?

The Nationwide Enterprise Analytics Office (formerly Customer Insights and Analytics) has more than 10 years of experience in end-to-end data product development and system integration. The culture to attract, train and develop talent, the technical advancement to apply the new methods, the model factory to productionalize models, and responsive processes to measure business impact have all contributed to positive business outcomes as well as this team’s fast growth. In this talk, we will introduce Nationwide’s data science capabilities through case studies of a few data products they have built and deployed.

As adoption of machine learning and artificial intelligence continues to grow, operational excellence in deployed ML assets will become as important as building the models themselves. The ability not just to develop models, but to version, deploy, and monitor them (and to automate this process) will drive differentiation and competitive advantage in the market. In this talk, we will discuss strategies for building and deploying high quality improvements to these models at scale, and how to engineer a model development lifecycle that supports this goal.

As personal data becomes abundant, the risk of sensitive data being leaked or misappropriated has become much greater. This risk is greatly increased by the ability to augment publicly available data. This occurs, in part, because aggregation erodes privacy — the combination of disparate and seemingly minor bits of personal information can be used to infer sensitive personal attributes. Consequently, organizations seeking to maintain trust with their customers must have robust frameworks in place to preserve privacy not only with their own internal data but also when those sources are joined to external data.

Data scientists apply technology and expertise to reveal hidden patterns. Or to find patterns that don’t really exist. Smart data scientists learn how to avoid ‘stupid data mining tricks’. The same skills that help us construct beautiful models that ignore reality can also help us identify false patterns. We’ll discuss where things can go off the rails and how to stay on track. Topics include the nature of randomness and the science of cross-validation. Join us for an interactive discussion that can help you see the patterns behind the patterns.

Data scientists may often be thought of as lone individuals striving to understand the data laid before them. But what is the optimal environment for a data scientist? How do they work best? What do they even do, and how?

With a master’s degree in HR, I got my first job as an HR analyst in the states. Three years later, I am now a data scientist who builds products and tells stories with data everyday. People often ask me what it’s like to transition from HR to data science. Well, it is like a wizard coming from a Muggle family. I will never forget the life changing moment when I built my first predictive model using R. It was magic!