Stratified Random Sampling#

Stage 4 · 🧩 Sampling, Partitioning & Segmentation · Lesson 23 of 56 · intermediate

◀ Previous · Cross-Selling · Next · Linear Congruential Random Number Generator (LCG) ▶ · ↑ Section

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

Representative by design#

A sample is only useful if it resembles the population it is drawn from. Stratified random sampling guarantees that resemblance for the characteristics you care about, by sampling within subgroups rather than trusting chance to balance them. It is a workhorse of survey design and, in this course, of splitting data for modelling.

Strata#

The method starts by dividing the population into strata — mutually exclusive, exhaustive subgroups that share a characteristic (gender, region, customer type). Strata are chosen to be internally homogeneous: alike within, different between. A simple random sample is then drawn independently from each stratum, and the pieces combined. Under proportional allocation, each stratum contributes in proportion to its share of the population, so the sample mirrors the whole.

Why not simple random?#

Plain random sampling can, by luck, under-represent a small but important group — draw 100 customers at random and a rare segment might barely appear. Stratifying removes that luck: every subgroup is present by construction, in the right proportion. The result is greater precision (lower sampling variability) than a simple random sample of the same size, especially when the strata differ from one another.

In machine learning#

The same idea is essential when splitting data. A stratified train/test split keeps the class proportions identical in both parts — vital for imbalanced problems, where a naive split might leave too few positive cases in the test set. In scikit-learn it is one argument: train_test_split(..., stratify=y), or StratifiedKFold for cross-validation. The next lessons need this discipline, because honest model evaluation depends on representative partitions.

Hint