Quality Audit Dataset
Upload a CSV or JSON file (max 5MB) and get quality metrics per column: missing values, duplicates, IQR outliers, cardinality, score 0-100, and recommendations. Fully client-sided - no data sent to the server.
Load Data Set
Drag your file here: CSV or JSON
or click to browse - max 5MB
How it works
Upload the file
Drag and drop a CSV (with headers) or JSON (an array of objects) file here. Up to 5 MB in size; fully processed on the client-side.
Automatic analysis
The tool infers column types, detects missing, duplicate, and outlier values, calculates the cardinality ratio.
Scoring and Remediation
Receive a score of 0-100, critical issues to prioritize and corrective actions to follow before loading the data set into production.
Come utilizzare Dataset Quality Audit
Load CSV or JSON file
Drag and select a .csv (with header) or .json (array of objects) file up to 5MB. The file stays in the browser: no upload to a server.
Column Automatic Analysis
The tool infers column types and calculates missing, duplicate, and outliers using IQR and cardinality ratio methods.
Read your score and critical issues
Receive a comprehensive score of 0-100, the list of critical priority issues and summary statistics (rows, columns, duplicates, processing time).
Apply Recommendations
Follow suggested corrective steps (e.g., handling missing values, removing duplicates) before loading the dataset into production or a machine learning pipeline.
Suggerimenti
- Perform an audit before every ML training: a dataset with too many missing values or unhandled outliers can skew the model's metrics.
- If a column has a cardinality ratio close to 1, check if it's an identifier to exclude from features rather than the training.
- Critical issues (labeled "CRITICAL") must be addressed first: they have a greater impact on the overall score than warnings.
Domande frequenti
How are outliers detected?
For numeric columns, the Interquartile Range (IQR) method is applied: a value is considered an outlier if it falls below Q1 - 1.5*IQR or above Q3 + 1.5*IQR, where IQR = Q3 - Q1.
What does cardinality mean in ratio?
It's the ratio of unique values to total rows in a column. A ratio close to 1 (≥0.95) indicates a nearly-identifying column (e.g., ID, email), useful for determining if a column can be a primary key or should be excluded from machine learning features.
How is the overall score of 0-100 calculated?
Score combines penalties for missing values (% by column), duplicate rows, outlier detection, and anomalous cardinality. High scores indicate a dataset ready for use, low scores signal issues to correct before analysis or training.
Are my data being uploaded to an external server?
No. All analysis (parsing, statistics, scoring) occurs 100% client-side via FileReader API: the file never leaves your browser.
What is the maximum file size I can analyze?
The limit is 5MB to ensure rapid analysis directly in the browser. For larger datasets, consider a preliminary sampling or use Python tools (pandas-profiling, ydata-profiling).