Dataset Profiler

What is Dataset Profiler?

Dataset Profiler is a browser-based utility for analyzing CSV data quality and structure before importing into spreadsheets, databases, BI tools, or ML pipelines. It runs entirely in your browser — no data is ever sent to a server.

Use it when you need to:

• Understand column data types — automatic inference of Text, Number, Date, Boolean
• Assess data completeness — missing value counts and percentages per column
• Measure cardinality — unique value counts and uniqueness ratios
• Get numeric summaries — min, max, average, median for numeric columns
• Spot data quality issues — automated observations for empty columns, mixed types, duplicate rows
• Export machine-readable profiles — JSON export for downstream tooling

Unlike server-based profilers, your CSV never leaves your device. This makes Dataset Profiler suitable for sensitive data, large files (limited only by browser memory), and offline use after the initial page load.

Key Concepts in Data Profiling

Cardinality (Unique Values)

Cardinality measures the number of distinct values in a column. A column with high cardinality (many unique values relative to row count) may be an identifier (e.g., user_id, email). Low cardinality suggests categorical data (e.g., status, country). The profiler reports both the absolute unique count and the unique percentage (unique values ÷ total rows × 100).

Type Inference

The profiler infers the dominant data type for each column by sampling non-empty values. Supported types: Text, Number (integer/float), Date (ISO 8601 and common formats), Boolean (true/false, yes/no, 1/0). When a column contains mixed types, it's flagged with a type confidence score and a quality observation warning.

Null Rates (Missing Data)

Null rate is the percentage of missing or empty values in a column. The profiler counts empty strings, whitespace-only cells, and common null representations (null, NA, N/A, -) as missing. Columns with >50% missing trigger a warning observation; >20% triggers an info observation. This helps you quickly identify sparse columns that may need imputation or removal.

Numeric Statistics

For columns inferred as numeric, the profiler computes: Min (minimum value), Max (maximum value), Average (arithmetic mean), Median (50th percentile). These are calculated only from valid numeric values — missing and non-numeric cells are excluded from the calculation.

How Dataset Profiling Works

Parsing and Delimiter Detection

The tool parses CSV text using a robust RFC 4180-compatible parser (shared with CSV Cleaner). It handles quoted fields, escaped quotes (""), and CRLF/LF line endings. Delimiter is auto-detected from comma, semicolon, tab, or pipe — or you can choose it manually.

Type Detection Algorithm

For each column, non-empty values are sampled and classified. A value is numeric if it parses as a finite number (integer or float). A value is a date if it matches ISO 8601 or common date patterns (MM/DD/YYYY, DD/MM/YYYY, YYYY-MM-DD). A value is boolean if it matches true/false, yes/no, 1/0 (case-insensitive). Everything else is text. The dominant type wins; if no type reaches >80% confidence, the column is flagged as mixed types.

Quality Observations Engine

After statistics are computed, a rule engine generates factual observations: High missing (>50%), Moderate missing (>20%), Empty column (0 non-empty), Mixed types (confidence <80%), Low uniqueness (<10% unique), Potential identifier (100% unique, non-empty), Empty rows (dataset-level), Duplicate rows (dataset-level). Observations are tagged warning or info severity.

Export Format

The JSON export contains three top-level keys: dataset (row/column counts, delimiter), columns (array of per-column stats objects with all metrics), and observations (array of quality findings). A meta.profiledAt timestamp is included. Your original CSV data is never included in the export.

Privacy

Your data stays in your browser. CSV files are processed locally using JavaScript and the FileReader API. No file contents are transmitted to any server. The tool works offline once the page is loaded. No analytics are collected on your CSV content.