Logistic Regression
Introduction
Logistic regression is a classification algorithm that predicts the probability of a binary outcome. Despite its name, it's used for classification, not regression. It uses the sigmoid function to map any input to a value between 0 and 1.
The Sigmoid Function
The sigmoid (logistic) function is the key to logistic regression:
σ(z) = 1 / (1 + e^(-z))
Properties:
- Output range: (0, 1)
- σ(0) = 0.5
- Smooth, differentiable
- Has a nice derivative: σ'(z) = σ(z)(1 - σ(z))
The sigmoid function maps any real value to (0, 1)
Interactive Logistic Regression
Click on the left (red region) to add class 0 points, or on the right (blue region) for class 1 points:
Blue curve: sigmoid function | Purple line: decision boundary (p = 0.5)
Controls
Model Parameters
Weight (w) = 0.000
Bias (b) = 0.000
Cross-Entropy Loss = 0.0000
Binary Cross-Entropy Loss
Logistic regression uses binary cross-entropy as its loss function:
L = -[y log(p) + (1-y) log(1-p)]
Where:
- y: true label (0 or 1)
- p: predicted probability
This loss function penalizes confident wrong predictions more heavily than uncertain predictions.
From Linear to Logistic
Linear Regression
- Predicts continuous values
- Output: y = wx + b
- Range: (-∞, +∞)
- Loss: Mean Squared Error
Logistic Regression
- Predicts probabilities
- Output: p = σ(wx + b)
- Range: (0, 1)
- Loss: Cross-Entropy
Key Takeaways
- Logistic regression is a linear classifier that outputs probabilities
- The sigmoid function maps linear combinations to probabilities
- Decision boundary occurs where the probability equals 0.5
- Cross-entropy loss is appropriate for probability predictions
- Can be extended to multi-class classification (softmax regression)
- Forms the basis for neural network output layers in classification tasks