Shreyas Kulkarni
Shreyas' Blog

Shreyas' Blog

Model Explainability

Photo by Thought Catalog on Unsplash

Model Explainability

Shreyas Kulkarni
·Jan 22, 2022·

3 min read

Subscribe to my newsletter and never miss my upcoming articles

Play this article

This week, I completed the course which I started last week on Coursera and I found a few new concepts that were foundational to the thought process of practicing data science, one of them is Explainability while building a model, at times if the model is giving accurate predictions but you are not able to figure out how the model is giving us the predictions stakeholders might not consider going with it.

When we look at predicted values or outcomes that are provided by our machine learning model, in many cases it's not enough to go with those predictions, we need to understand what is happening behind the curtains, we need to know what parameters the model is taking into account or if the model contains any bias.

at times, with machine learning models data goes in and predictions (output) come out but we don't know what exactly happens inside? and all you can say is nobody knows, it’s like a black box.

there comes Explainability, which means that you can explain what happens in your model from input to output. It makes models transparent and solves the black box problem.

Simply, Explainability is being able to quite literally explain what is happening.

Why is explainability important?

  1. Accountability: necessary to avoid similar problems in the future.

  2. Trust: for humans, it's hard to sink in something that they don't understand fully.

  3. Compliance: to ensure compliance with company policies, industry standards, and government regulations.

  4. Performance: explainability will help to understand where exactly fine-tune the model.

  5. Enhanced control: self-explanatory

Explainability is more important in high-risk domains like healthcare or finance. basically, tiny mistakes can lead to the death of patients or the loss of millions.

Two ways to approach explainability:

Globally – This is the overall explanation of model behavior.

Locally – This tells us about each instance and feature in the data individually.

Similarly, another important parameter that I learned about is that different countries have different laws and regulations regarding how companies process user data, different types of consents that are being explicitly required from the user or the respective governments.

However, I came across this site which helps us to understand laws in different countries(related to data) with simple visuals and also in text format as well.

image.png DATA PROTECTION LAWS OF THE WORLD

You can simply notice that some countries have heavily defined laws and regulations around data and some of them don't even have laws and regulations themselves. most of them are limited and moderate as well. moreover, you can even compare your country with others and gain some insights to satisfy your curiosity.

That's all for this week, something interesting next week, and you can always connect with me Here😃

 
Share this