Machine Learning at Employment and Social Development Canada
We have used machine learning to find the best comparison cases via classification and to find distinct groups of participants with varying needs among the most disadvantaged youth.
Employment and Social Development Canada uses innovative machine learning in their Evaluation Directorate.
Contributor
Tristan Rikhi
Evaluation Officer
ESDC
Transcript
Hello, my name is Tristan Rikhi and I am an analyst with the Evaluation Directorate at Employment and Social Development Canada. I'm excited to share with you some of the innovative ways that we have been using machine learning in the Directorate.
For the previous Youth Employment Strategy Evaluation, we have used machine learning to find the best comparison cases via classification and to find distinct groups of participants with varying needs among the most disadvantaged youth. The newest technique that we are working on combines econometric methods with the predictive power of machine learning to produce individualized treatment effects which have program impacts that are specific to each participant.
Decomposing our single average treatment effect into the new individualized treatment effects affords an unprecedented level of granularity for causal influence, thereby allowing evaluators to determine optimally which programs and services are most effective and for whom.
Some of the challenges we face are: One, it is a relatively new and very technical approach, so it takes very specialized knowledge to implement successfully, and Two it can be computationally very expensive.
To overcome the first challenge, we are working with Professor Michael Lechner, a leading expert in labour market evaluations and was one of the pioneers for this specific technique we are using to guide our analysis and provide helpful feedback along the way.
With the help of senior management, we have been able to acquire the tools such as a powerful hardware that we need to perform this type of analysis. Finally, we are amongst the first countries in the OECD who are exploring these new methods for impact evaluation and we look forward to sharing our results when available. Thank you.
Courses/ Resources
Discover Series: Discover Data