Thread by @pvergadia

Day 13 #31DaysofML

⚖️ How to deal with imbalanced datasets?⚖️
Most real-world datasets are not perfectly balanced. If 90% of your dataset belongs to one class, & only 10% to the other, how can you prevent your model from predicting the majority class 90% of the time?

🧵 👇

🐱🐱🐱🐱🐱🐱🐱🐱🐱🐶 (90:10)
💳 💳 💳 💳 💳 💳 💳 💳 💳 ⚠️ (90:10)
There can be many reasons for imbalanced data. First step is to see if it's possible to collect more data. If you're working with all the data that's available, these 👇 techniques can help

#31DaysofML 2/7

Here are 3 techniques for addressing data imbalance. You can use just one of these or all of them together:
⚖️ Downsampling
⚖️ Upsampling
⚖️ Weighted classes

#31DaysofML 3/7

📌 Downsampling 📌
This technique removes a random subset of the majority class from your data
Original dataset: 🐱🐱🐱🐱🐱🐱🐱🐱🐱🐱 🐶
Downsampling: 🐱🐱🐱🐱 🐶
Note: requires starting with a big enough dataset to not lose any meaning present in the data

#31DaysofML 4/7

📌 Upsampling 📌
In this technique you generate new examples for your minority class that are in a similar feature space as existing examples
Original dataset: ⚽️⚽️⚽️⚽️⚽️⚽️⚽️ 🏀
Upsampling: ⚽️⚽️⚽️⚽️⚽️⚽️⚽️🏀🏀🏀🏀🏀

#31DaysofML 5/7

📌 Weighted classes📌
In this technique you tell your model to give more weight to specific examples in your dataset
Original dataset: 🍎🍎🍎🍎🍎🍎🍎 🍋
Pay more attention to 🍋 minority class

#31DaysofML 6/7

This tip comes from my ML mentor @SRobTweets who recently co-authored a book on Machine Learning Design Patterns along with @lak_gcp & Michael Munn

Here's the link to the book 👉 amzn.to/3d63GO4
SMOTE paper 👉 arxiv.org/abs/1106.1813

#31DaysofML 7/7