Numeric vs categorical columns

Every cleaning decision you make in Dotwave starts with one question the tool answers for you automatically: is this column numeric or categorical? The two kinds of column carry different information, need different statistics to describe them, and call for completely different fixes. Understanding how Dotwave draws the line — and what it measures on each side of it — lets you read the profile the way the tool intends and choose the right action in every later step.

How Dotwave classifies a column

When it profiles a file, Dotwave inspects the values in each column and assigns it to one of two families. A column of quantities you could meaningfully add, average, or plot on a number line — revenue, quantity, age, temperature — is treated as numeric. A column of labels that name a category rather than measure an amount — country, product line, status, department — is treated as categorical, even when those labels happen to be stored as numbers, as with a postal code or a region ID. The distinction is about meaning, not storage: averaging a set of postal codes produces nonsense, so Dotwave keeps such fields on the categorical side where counting, not arithmetic, is the sensible operation.

Statistics for numeric columns

For every numeric column, Dotwave computes a full set of descriptive statistics so you can judge the column's centre, spread, and shape at a glance:

• Mean — the arithmetic average, the balance point of the values.
• Median — the middle value when the data is sorted, unaffected by extreme outliers.
• Std — the standard deviation, a measure of how widely the values spread around the mean.
• Min and Max — the smallest and largest values, which quickly expose impossible entries such as a negative age.
• Q1 and Q3 — the first and third quartiles, marking the boundaries of the middle 50% of the data.
• Outlier count — how many values fall far enough outside the interquartile range to be flagged as unusual.

Read the mean against the median: when they are close, the column is roughly symmetric and the mean is a fair summary. When they diverge, the column is skewed and the median is the more trustworthy centre — which is exactly why Dotwave leans on the median when it suggests fills.

Statistics for categorical columns

Categorical columns cannot be averaged, so Dotwave describes them by counting instead:

• Unique count — how many distinct labels appear in the column.
• Top value — the label that occurs most often.
• Frequency — how many rows carry that top value, which tells you how dominant it is.
• Sample values — a handful of actual entries so you can eyeball the formatting and spot inconsistent spellings or stray whitespace.

The sample values are deceptively useful: seeing London, london, and London side by side is often the first sign that a column needs whitespace trimming or a merge of look-alike categories.

The high-cardinality warning

When a categorical column contains more than 20 unique values, Dotwave raises a high-cardinality warning. High cardinality is a signal worth pausing on. Sometimes it is legitimate — a customer name or order ID naturally has as many distinct values as there are records. But often it points to a column that has not been standardized, where genuine categories are fragmented across dozens of near-duplicate spellings. A "Status" field with 40 unique values almost certainly hides a much smaller set of real statuses that inconsistent data entry has splintered. High cardinality also matters downstream: one-hot encoding such a column would explode it into dozens of new fields, so the warning is your cue to clean or reconsider before you encode.

Why the distinction drives cleaning decisions

The numeric-versus-categorical split is not academic — it dictates which fixes are even available to a column. A missing numeric value can be filled with a mean or median; a missing categorical value cannot, so Dotwave offers the mode instead. Outlier capping only makes sense on a number line, so it is a numeric-only operation. Merging look-alike categories and label or one-hot encoding belong exclusively to categorical columns. Because the classification determines the menu of suggested actions you see in Steps 2 through 4, getting each column into the right family — and correcting any that were misread on upload — is the quiet foundation that makes every later suggestion sensible.