Correlation Matrix Calculator
Computes the Pearson correlation coefficient — a measure of linear relationship strength and direction, ranging from -1 (perfect inverse relationship) to +1 (perfect direct relationship) — between every pair of numeric columns in your dataset, displayed as a color-coded matrix.
Correlation measures a linear relationship specifically — two variables can have a strong non-linear relationship (like a U-shape) while still showing a near-zero Pearson correlation, since the metric doesn't capture non-linear patterns.
Reading the matrix
The diagonal is always 1: every column correlates perfectly with itself. The matrix is symmetric, so only half of it carries information — the top-right and bottom-left mirror each other.
Rough interpretation of the magnitude, ignoring sign: below 0.3 is weak, 0.3 to 0.7 is moderate, above 0.7 is strong. These are conventions rather than thresholds, and what counts as strong depends heavily on the field — 0.4 is a substantial effect in social science and noise in a physics experiment.
The uses that actually matter
Finding redundant columns. Two features correlating above about 0.9 carry nearly the same information. In a regression this is multicollinearity: the model still predicts well, but the individual coefficients become unstable and their interpretation unreliable. Dropping one is usually the right call.
Spotting leakage. A feature correlating almost perfectly with your target is often not a feature at all but a restatement of the answer — a "discount applied" column that only exists once a sale completed, for example. Catching this before training saves a model that looks impeccable in validation and fails in production.
Orienting yourself in unfamiliar data. The matrix shows, in one view, which measurements move together — which is the fastest way to form a hypothesis about what the dataset is describing.
What Pearson misses
Pearson's coefficient measures linear association only. A perfect parabola scores near zero. It is also sensitive to outliers, and a single extreme row can dominate the value in a small dataset. Where a relationship is monotonic but curved, Spearman's rank correlation is the better measure — and where it matters, look at the scatter plot before trusting any coefficient.