Theme 1: Fraud Detection

Imagine you work for Travelers Insurance Company's fraud detection department as a modeler. Your colleagues, who are unfamiliar with statistics, would like you to create a predictive model based on historical insurance claim data. Your team is concerned about fraud detection accuracy as well as the key drivers that cause fraudulence. For this case competition, your group is tasked with identifying first-party physical damage fraudulence and explaining the indicators of fraudulent claims.

For more details about this theme, please register as a team or register to join a team for the Statathon, and we will send you a link to work on this challenge through Kaggle. The top 5 teams will be invited to give a virtual presentation of their solution (15 minutes) and answer questions from the judges, who will determine the winning teams.

(Data sets are synthetic, provided by Travelers)

More information available here

Theme 2: Did the Customer Take Action after the Alert?

HSB offers equipment breakdown insurance and provides Internet of Things (IoT) sensor solutions to help customers avoid or reduce damage to their building, equipment, or machinery. These sensors monitor essential health and performance variables that alert customers to take corrective action before critical system failure occurs.

Using these sensors, HSB has created a customer alert program in which an alert (in the form of text, email, etc.) is sent to a customer upon detection of abnormal sensor activity. In order to determine the success and usefulness of this alert program, HSB must record whether or not a customer took corrective action after receiving an alert. Unfortunately, direct follow-up with all customers is infeasible due to the large number of deployed sensors as well as potential customer unresponsiveness, thus an automated system is needed to deduce if a customer took appreciable corrective action.

In this challenge, your team will be given time series data of a custom system health metric, which consists of 25 total alerts. Using whatever methodology you see fit, develop a model that determines if an insured took action to mitigate poor operating conditions after receiving an alert.

For a full description of the dataset, please see this link

A brief data tutorial can also be found here