CSV Column Analyser — Free Online CSV Statistics & Data Profiler

CSV Column Analyser Guide — Data Profiling and Descriptive Statistics

Data profiling is the process of examining a dataset to understand its structure, content, and quality before using it for analysis, reporting, or machine learning. Column-level analysis — examining each column in a CSV file individually — is the most important step in any data profiling workflow. It reveals which columns have missing values, which contain outliers, what the typical ranges and distributions are, and where data quality issues lurk that could corrupt downstream analysis.

What is exploratory data analysis (EDA) on CSV files?

Exploratory data analysis (EDA) is the practice of summarising and visualising a dataset’s key characteristics before formal modelling or analysis. For CSV files, EDA typically starts with column-level statistics: count, null rate, minimum, maximum, mean, and median for numeric columns; unique value count and most frequent values for text columns. These statistics form a data profile that guides decisions about data cleaning, feature engineering, and modelling strategy. The LazyTools CSV Column Analyser automates this entire first-pass EDA in seconds, eliminating the need to write Python pandas code or open a Jupyter notebook for initial data inspection.

How to identify missing data in a CSV file

Missing data appears as empty cells in a CSV file — cells where a value should exist but was not recorded, not collected, or was lost during export. The null count and null percentage for each column directly quantify missing data. A column with 30% null values should raise immediate questions: was data collection optional for that field? Were records incomplete at source? Is the null pattern random (MCAR), related to other variables (MAR), or systematic (MNAR)? The answer determines the appropriate imputation strategy. The data quality score in the LazyTools analyser (100 minus null %) gives an at-a-glance health check: a score below 70 typically means the column requires cleaning or imputation before use in analysis.

Understanding mean vs median in CSV data analysis

The mean and median are both measures of central tendency — the typical or central value in a distribution — but they respond very differently to skewed data and outliers. The mean (average) sums all values and divides by the count. One extreme value can shift the mean dramatically. For example, in a salary dataset where most employees earn $40,000–$80,000 but one executive earns $2,000,000, the mean is pulled far above the actual typical salary. The median, the middle value in a sorted list, is unaffected by this single outlier and gives a much better picture of the typical salary. When mean and median differ significantly in your CSV column analysis, it is a strong signal that the distribution is skewed or contains outliers that require investigation.

How to detect outliers in a CSV column

The LazyTools CSV Column Analyser uses the two-standard-deviation rule for outlier detection: any value more than 2 standard deviations above or below the mean is flagged as an outlier. For a normal distribution, approximately 95.4% of values fall within 2 standard deviations of the mean, so about 4.6% of values would be flagged even in perfectly clean data. In practice, an outlier count significantly higher than 5% of the column’s non-null values warrants investigation. Outliers can represent data entry errors (a salary entered as 500000 instead of 50000), measurement errors (a sensor recording 9999 when it fails), or genuinely unusual cases that are valid but need separate treatment in modelling.

Data quality scoring for CSV columns

Data quality is multi-dimensional — it encompasses completeness (are all values present?), consistency (are values in the expected format?), accuracy (do values reflect reality?), uniqueness (are there unexpected duplicates?), and validity (do values fall within expected ranges?). The LazyTools quality score focuses on completeness — the proportion of non-null values — as it is the most universally applicable and directly measurable dimension from raw CSV data. A score of 100 means every row has a value in that column. A score of 80 means 20% of rows are missing data. Combine the quality score with unique count (to detect duplicate-heavy columns) and the outlier count (to assess range validity) for a fuller picture of column quality.

Using CSV statistics for data cleaning decisions

Column statistics directly inform data cleaning decisions. A column with 90%+ null values may be safe to drop from the dataset. A numeric column where mean and median differ by more than 50% likely has outliers that need capping or removal before regression modelling. A text column with only 5 unique values across 10,000 rows is a categorical variable that should be one-hot encoded or label-encoded for machine learning. A column where the minimum value is negative in a dataset that should only contain positive quantities (e.g. order quantities) signals a data entry or extraction error. A column with 100% unique values across all rows is a candidate for a primary key or identifier field. All of these decisions start with the column statistics the LazyTools analyser computes.

Exporting CSV analysis reports for documentation

Column analysis reports are valuable documentation assets in data engineering and data science projects. A data dictionary — a structured document describing each column in a dataset, its type, typical range, null rate, and unique value count — is a standard deliverable in enterprise data projects. The LazyTools Export Report CSV function generates a machine-readable version of the analysis results: one row per column, all statistics as columns. This can be imported into Excel, appended to a project wiki, or used to generate automated data quality alerts. The Copy Markdown Table function produces a formatted markdown table ready to paste into GitHub READMEs, Confluence pages, Notion databases, or any Markdown-aware documentation system.