Learn what logistic regression is, then train it in your browser on a sample dataset or your own CSV file.
Logistic regression is a classification model. Instead of predicting a number, it predicts the probability that an observation belongs to the positive class (label = 1).
It first computes a linear score (the log-odds) from the features:
z = b0 + b1·x1 + b2·x2 + ... + bn·xn
then squashes it into the range (0, 1) with the sigmoid function:
p = σ(z) = 1 / (1 + e^(−z))
A sample is predicted positive when p ≥ threshold (default 0.5).
Coefficients are fitted by minimising the log loss (binary cross-entropy):
L = −(1/N) Σ [ y·log(p) + (1−y)·log(1−p) ] + λ·Σ b²
using gradient descent. The optional L2 penalty (λ) shrinks coefficients and reduces overfitting.
Each coefficient is a change in log-odds. exp(b) is the odds ratio: how many times the odds of the positive class multiply when that feature increases by one unit (one standard deviation if you standardise).
Use it when you need a fast, interpretable baseline for a binary outcome (disease / no disease, pass / fail, churn / stay), when you care about calibrated probabilities, or when you must explain the effect of each variable. It assumes a roughly linear relationship between the features and the log-odds; for strongly non-linear boundaries prefer trees, boosting or kNN.
The sigmoid curve: linear score z on the x-axis, predicted probability on the y-axis. The dashed line is your decision threshold.
Loads instantly, no upload needed.
First row = column names. Target column must have exactly two distinct values.
Accuracy — fraction of correct predictions. Precision — of predicted positives, how many were right. Recall (sensitivity) — of real positives, how many were found. Specificity — of real negatives, how many were correctly rejected. F1 — harmonic mean of precision and recall. AUC — probability that a random positive scores higher than a random negative (0.5 = random, 1.0 = perfect).