Machine learning models, gradient-descent optimisers, and distance-based algorithms like k-NN and k-means all treat raw numeric magnitude as meaningful signal. A feature measured in thousands (house prices like 120, 250, 85, 340, 175, 95, 420, 210) will silently dominate a feature measured in single digits unless the two are brought onto the same scale first. Data Normalizer handles both standard rescaling strategies in one place: Min-Max Scaling maps every value into any target range you choose, and Z-Score Standardisation shifts the dataset to zero mean and unit variance.

Paste any delimiter-separated list of numbers (commas, spaces, semicolons, pipes, tabs, or newlines all work), set your method and target range, and the tool produces the normalised values, a value-by-value breakdown table with a proportional bar for each row, and a downloadable CSV you can feed straight into pandas, R, or a spreadsheet.

Min-Max vs Z-Score: Choosing the Right Method

Min-Max Scaling guarantees every output lands in your target range (default [0, 1]). Use it when the algorithm expects bounded inputs: neural network activation functions, image pixel values, or any tool that interprets 0 and 1 as off and on. The formula is x' = (x − min) / (max − min), extended to a custom range as x' = targetMin + (x − min) / (max − min) × (targetMax − targetMin). One non-obvious consequence: a single outlier compresses all other values near one end of the range. The sensor-reading sample (10, 12, 11, 13, 10, 99, 11, 12) demonstrates this — the outlier maps to 1.0 and every other value falls below 0.04.

Z-Score Standardisation does not bound the output; it re-centres the data around 0 and scales by standard deviation. Use it for linear regression, PCA, and any algorithm that assumes normally-distributed inputs. The formula is x' = (x − μ) / σ where σ is the sample standard deviation (denominator n−1). A z-score of +2 means the value sits two standard deviations above the mean; −1.5 means one-and-a-half below. When all values are identical (σ = 0), every output is 0 rather than undefined, so the tool returns a result instead of an error.

Worked Example

Load the lead sample: eight house prices in thousands — 120, 250, 85, 340, 175, 95, 420, 210. With Min-Max Scaling and the default [0, 1] target range, the minimum (85) maps to 0.0 and the maximum (420) maps to 1.0. The value 175 maps to (175 − 85) / (420 − 85) = 90 / 335 ≈ 0.268657. Switch to Z-Score: the sample mean is 211.875 and the sample standard deviation is approximately 114.27, so 420 normalises to (420 − 211.875) / 114.27 ≈ 1.821, about 1.8 standard deviations above average.

Accuracy, Limits, and Boundary Behaviour

• Minimum 2 values required. A single value produces the validation message "At least 2 values are required to normalize a dataset." — a one-element set has no spread to scale.
• Negative inputs work correctly. The stock-returns sample (−2.3, 1.4, 0.8, −0.5, 3.1, −1.2, 0.3, 2.7) normalises without special handling; Min-Max maps −2.3 to 0.0 and 3.1 to 1.0.
• All-equal datasets. Min-Max maps every value to the midpoint of the target range rather than dividing by zero. Z-Score returns 0 for all values when σ = 0.
• No ceiling on value count. The table renders all rows with alternating row shading; the visual bar column is hidden on small screens to keep the layout readable.
• CSV export preserves the full precision set by the Decimal Places selector (2–10 places). The Copy CSV button and the Download CSV button produce the same content: two columns, original and normalized, one row per input value.