𝗗𝗶𝘀𝗰𝗿𝗶𝗺𝗶𝗻𝗮𝘁𝗶𝘃𝗲 vs. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 Models – A simple and intuitive explanation!!!

𝗗𝗶𝘀𝗰𝗿𝗶𝗺𝗶𝗻𝗮𝘁𝗶𝘃𝗲 vs. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 Models – A simple and intuitive explanation!!!

Imagine you and your friend work at a bank.

Your job is to detect fraudulent Transactions. You are expert in identifying whether a specific transaction is fraudulent or not based on patterns in past transactions.

Your friend’s job is to observe customer spend. He is expert at modeling customer spending habits to understand their typical purchase patterns, preferences, and financial behavior.

You are a discriminative Model while your friend is a generative model.

How?

You focus on the decision boundary, distinguishing between classes (fraudulent vs. legitimate).

While your friend understands the data distribution, enabling it to generate new, realistic examples.

That’s it.

Let’s dig deeper.

⭕ Discriminative Models

Discriminative models are all about distinguishing between classes. They model the decision boundary that separates different categories without trying to understand the underlying data distribution.

Common Examples:
• Logistic Regression
• Support Vector Machines (SVMs)
• KNN and many more..

⭕ Generative Models

Generative models aim to capture the data distribution, allowing them to not only classify but also generate new examples from the same distribution.

Why They’re Useful:
• Generating realistic images (e.g., GANs generating human faces).
• Filling in missing information (e.g., generating text in ChatGPT).
• Unsupervised learning (e.g., learning clusters without labeled data).

Common Examples:

• Naive Bayes
• Generative Adversarial Networks (GANs)
• Variational Autoencoders (VAEs)

When to Use What?

• Use Discriminative Models:
When you need accurate predictions for tasks like spam filtering or image classification.
• Use Generative Models:
When you want to generate new data or model complex relationships.

PS: My new cohort of Master DataScience Program is starting on Nov 28. DM for more details.