Percentiles and Quantiles
What you'll learn: How percentiles and quantiles pinpoint specific positions within your data distribution, helping you understand where individual values stand relative to the whole dataset.
What Are Percentiles and Quantiles?
Imagine 100 students lined up by test score from lowest to highest. The student at position 75 scored better than 75% of the class—that's the 75th percentile. Percentiles divide your data into 100 equal parts, telling you what percentage of observations fall below a given value.
Quantiles are the general term for these division points. They split data into equal-sized groups:
- Quartiles divide data into 4 parts (25%, 50%, 75%)
- Percentiles divide data into 100 parts (1%, 2%, ..., 99%)
- Deciles divide data into 10 parts (10%, 20%, ..., 90%)
Why They Matter in Machine Learning
Unlike mean and standard deviation (which you learned about in central tendency and spread), percentiles don't get thrown off by extreme outliers. They show you the actual distribution shape:
- The 50th percentile (also called the median) is the middle value
- The 25th percentile (Q1) and 75th percentile (Q3) mark the "middle half" of your data
- The interquartile range (IQR = Q3 - Q1) measures spread without being affected by outliers
Think of percentiles as "what value separates the bottom X% from the rest?" This helps you identify unusual data points, understand skewness, and make decisions like "which features have values in the extreme 1%?"
Key Takeaway: Percentiles and quantiles reveal where specific values sit within your data's distribution, providing robust position measures that resist the influence of outliers—essential for understanding real-world ML datasets.