Date: 2/6-2021
Time: 09:00-12:00 CET (GMT +1)
Place: Microsoft Teams.
Agenda
- Fit statistics & what to use when
- Validating your models is a crucial step of the modelling process. As we develop more and more methods of measuring a models goodness of fit, we are challenged to use the correct statistic for the correct model.
- In this presentation we will go through a handful of these fit statistics and explain where and why certain different fit statistics should be used.
- We aim to answer the question “Why are there so many and how do they differ?”
- By Philip Sierpinski, SAS
- Interpretability
- Today a data scientist has a lot of different algorithms in his/her toolbox.
- Some of the algorithms generate very complex models which we often describe as “Black Boxes”, they are often very accurate, but can we understand them?
- Are the models: Fair, Accountable Transparent Trustworthy Explainable?
- In this session we are covering how we can open up the black box and add post analysis method, to help us to understand the model, such as:
- Partial Dependence (PD),
- Individual Conditional Expectation (ICE),
- Local Interpretable Model-agnostic Explanations etc.
- By Mathias Lanner, SAS
- Data ethics
- The way you treat data during you model development has a big influence on the validity of the model outputs. In this presentation we will discuss some of the pitfalls that exists when treating data during model development and how we can mitigate them. Along the way we will among other discuss why a Wall Street investment bank does not allow work related discussions between colleagues and why some statistics professors find it wrong to plot data before model estimation.
- By Rune Nielsen, SAS
- Model Evaluation Metrics in Machine Learning. Understand the problem and a practical approach
- In my role as a model manager for a model factory for 1-1 marketing campaigns, we defined some metrics to evaluate models.
- Our goal is to choose the best (good) performing models that will stay stable over time, as we don’t want to change the models every month.
- Every model is for an individual business case by itself.
- By Hans de Witt, Telenor
Find the presentations below
Click on the download symbol to the right of the PDF filename to download the PDF file.
If you would like to access a recording of one of the presentations, please send a mail to frans.holm@sas.com. We will then sent you a mail with a link to the recording.
Subscribe to Nordic SAS User Newsletter (monthly)
Back to SAS Nordic User site
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.