Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail Machine Learning – The Value add to Our Services Christina Hoy VP, Corporate Business Information and Analytics October 5, 2016
15
Embed
WSIB presentation at the Chief Analytics Officer, Fall 2016
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Machine Learning – The Value add to
Our Services
Christina Hoy
VP, Corporate Business Information and Analytics October 5, 2016
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
WSIB – Who we are
2
A “hybrid” – independent trust agency in
Ontario, Canada:
- legislated by the Ontario government and
responsible for administering the
Workplace Safety and Insurance Act
(WSIA)
- funded by the employers of Ontario
Administers compensation and no-fault
insurance for Ontario workplaces
Serves workers and employers ranging from
small businesses to large private- and public-
sector organizations Source: By the Numbers: 2015 WSIB Statistical Report www.wsibstatistics.ca
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Then and Now – Strategic Change at WSIB
3
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Defining Machine Learning
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Machine Learning at WSIB
5
Operationalizing Machine Learning can transform WSIB, moving beyond monitoring, to delivering new value added services
Machine learning means strategically investing in advanced analytics tools that add value to WSIB. Machine Learning will enable WSIB to:
– Bring effective innovative solutions to better serve our customers and improve outcomes,
– Improve decision making in real-time by embedding machine learning into business decisions made at the front-lines of our
operations,
– Build a culture of continuous improvement by implementing a feedback loop between decision makers and senior management, and
– Scale up the use of data by pulling together information from a number of disparate sources to create models that better target customer segments and geographies to produce highly specific products/services for our customers.
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Machine Learning
6
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Examples of Machine Learning at WSIB
The following are early examples
of machine learning:
■ Claims Risk Scoring
■ Fraud Detection
– Employer Premium
– Service Provider
7
Machine Learning at WSIB implements a continuous feed back loop
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Claims Risk Scoring Current:
■ Claims risk scoring answers the question; how do changes to different factors influence the outcome of a claim at a specific time after injury?
■ In order to address this question our data scientists worked collaboratively with senior management to understand the business issue.
■ Once the issue was understood, a review of the available data took place.
■ We then used SAS tools to bring this data together from a number of disparate sources.
■ Finally, a model was created and supervised machine learning techniques were applied.
■ The business reviewed the outputs of this model for acceptance and we then
successfully deployed to middle management for use in prioritizing
the review of claims.
Future:
■ Our plan is to deploy this model to front-line decision makers by
integrating directly into our front-line systems.
■ By doing so we hope to implement a decision
management system for feedback loop to assess
assess the impact of decisions made.
8
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Claim Risk Scoring Example
Joe is a 45 year old male. He is currently employed in the Forestry industry
and is an equipment operator.
Joe is a French speaking Canadian who has never
had a prior claim.
Joe works for a private company with less than 20
employees and makes approx. $500 per week.
Last week Joe fell at work and was injured.
Back Sprain
Shoulder Leg fracture
8
Probability of staying on benefits changes for different injury types
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Fraud Detection
10
Current: ■ How do we better understand and identify fraud, waste and abuse.
■ To do so we created two fraud detection models to use as a targeting tool for high risk employers and service provides:
Employer Premium Reporting, and
Service Provider.
■ Once we understood the business issue we reviewed both internal and external data and combined them using specialized software.
■ A model was then created and we applied both unsupervised
and supervised machine learning techniques.
■ The outputs of this model were then reviewed by
our front-line business for acceptance, at which
point we deployed in the form of reports
and target lists.
Future:
■ As with our older model we plan to deploy this to
front-line decision makers which will enable a more
robust feedback loop on decisions.
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
How did we embed this into our operations?
11
Formal policies,
systems and practices
Informal practices, and
symbolic actions
Belief, values and
attitudes
The key to success is to create an Analytics Driven Culture
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Building the right team
12
At least four different skill streams are required:
Design the
necessary infrastructure
Utilizing data
produced by data science
Wrangling data
from data marts and warehouses
Constructs, designs
or arranges usable data
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Continuous Improvement: the Machine Learning Journey
■ Continue culture shift among senior management by building trust through small
scale focused initiatives with high potential successes.
■ Embed machine learning by implementing these models
into the tools that front-line decision makers use.
■ Fill technological gaps by strategically investing
in new tools and products that enable the use
of machine learning techniques more
effectively.
■ Continue to build in invest in our Information
Management Architecture.
13
WSIB will strive to create an environment of analytics throughout the
organization to spur creativity in how we solve seen and unseen problems
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail
Our Journey Continued…
■ Continual improvement of data quality and access by implementing a data Quality
Assurance Program.
■ Change in team structure to ensure our workforce has
the right capability to meet the needs of a data
driven future.
■ Drive innovation in data analytics by aligning
a strategic team of experts with a common
focus on maximizing business objectives
through a Centre of Excellence.
14
WSIB will strive to create an environment of analytics throughout the
organization to spur creativity in how we solve seen and unseen problems
Workplace Safety and Insurance Board | Commission de la sécurité professionnelle et de l’assurance contre les accidents du travail