Explain like I'm five · AI Theory

What is Logistic Regression?

Logistic regression, told as a story about feeding the scale's number through an S-curve to get a pass probability.

the sorter has to answer pass or fail — only one of the twothe scale gives a number… but which side is it?

The sorter has to answer pass or fail — only one of the two.The scale gives a number, but the number by itself isn't the answer.

a score of 130%? a probability can't go past 100a score of -40%? a probability can't go negativeusing the scale's raw number as a probability as-is gives values outside the 0-to-1 range

Using the raw score directly as a probability makes no sense.A probability has to stay between 0 and 1 — but a raw score doesn't respect that range.

logistic regression feeds the scale's number through an S-curve, turning it into a pass probability between 0 and 1the scale's number (score)0.5the S-curve (sigmoid)

Logistic regression feeds the scale's number through an S-curve to get a pass probability.After the S-curve, no matter how big or small the score, the result always lands between 0 and 1.

Score itweighting and adding features into one score
Squash it with an Ssqueezing the score between 0 and 1
Pass the cutoff, call it PASS0.5 is the usual cutoff line
Draw the boundarythe line that splits pass from fail
LINEAR REGRESSION vs LOGISTIC REGRESSIONlinear regression — predicts the raw number itself (continuous)logistic regression — predicts a pass probability (0 to 1)linear regression — e.g. predicting produce weightlogistic regression — e.g. classifying pass vs. fail

Linear and logistic regression both use the scale — but they answer different questions.Linear regression asks 'how much'; logistic regression asks 'which side'.

0.5the pass/fail boundary is still one straight linewithout transformed features, it can't draw a curved linelogistic regression only squashes the output — underneath, it's still one straight dividing line

Logistic regression squashes the score with an S — but underneath, one straight line still does the dividing.For a more complicated boundary, you'd check performance with a confusion matrix and reach for a different method.

In one breath

Logistic regression = feeding the scale's score through an S-curve to get a pass probability between 0 and 1.

Logistic regression is a supervised classification method that passes a linear combination (weighted sum) of input features through a sigmoid function, turning the result into a probability between 0 and 1. If that probability clears a threshold (usually 0.5), it's classified as the positive class; otherwise, the negative class. Despite the word 'regression' in its name, it's actually used for classification — and because its decision boundary is a straight line (a hyperplane) in feature space, it struggles, without transformed features, on data that needs a curved boundary.

When grown-ups say it

Logistic regression
Feeding the scale's number through an S-curve to get a probability. Despite its name, it's used for classification.
Sigmoid function
The S-curve that squashes any number into 0 to 1. The core mechanism of logistic regression.
Linear regression
Predicting the raw scale number itself — a continuous value. The original owner of the scoring method logistic regression borrows.
Decision boundary
The line that splits pass from fail. In logistic regression, without transformed features, this boundary is a straight line (or flat plane).
Threshold
The cutoff that turns a probability into pass or fail. Usually 0.5, but it can be moved depending on the situation.
Odds ratio
How many times the odds (pass probability ÷ fail probability) change when a feature changes. The term used when interpreting a logistic regression coefficient.
Confusion matrix
The ledger that splits pass/fail calls into four boxes. Logistic regression's own performance is checked the same way. → The Four-Box Stamp Ledger
Clustering
Finding groups on its own, with no answer labels. A neighboring method that starts from the opposite assumption. → Grouping by Likeness, With No Name Tags
Next story
The Produce That Stands Out from the Crowd →
Want another term as a picture book? Request →