PCA Explained in Simple Words | Machine Learning Made Easy

Key Takeaways

Ever wonder what PCA is? Let's break it down. PCA stands for principal component analysis. It's a dimensionality reduction technique that takes highdimensional data and projected into fewer dimensions while keeping as much variation or information as possible. Think of it like this. You have a data set with hundreds of columns. PCA finds out the best angles to view your data. It's like choosing the best camera angle for a group photo such that everyone is clearly visible and spread out. By directions we mean straight line through the cloud of data points that captures the most variation in the data. These best direction are called principal components. The first principal component is the direction where your data varies the most like the longest stretch of your data cloud. The second principal component captures the next biggest variation. But here's the catch. It must be orthogonal at a right angle to the first. This ensure it captures new information not just rotated view of the same spread. more components can follow each orthogonal to the ones before it. Each capturing less and less variation. But why to use PCA? Because it cuts complexity. For example, from 100 features to 10 features. Then it removes redundancy by combining correlated features. Next, it enables 2D 3D visualization of patterns. It reduces noises by ignoring small random variation and it may also reduce overfitting by keeping only key uncorrelated features. In short, PCA is a powerful pre-processing and exploration tool in machine learning pipeline. It simplifies your data set while keeping the essence of the information. If you found this helpful, hit the like, share, and subscribe button and follow for more a IML concept explained simply.