Clean Your Dataset Before Training a Machine Learning Model

Most model problems start in the data. A short cleaning routine catches the common ones early.

Open the Data Profiler

Duplicate rows, blank cells, mixed date formats and numbers stored as text all quietly hurt a model. Duplicates can leak between training and test sets and make a score look better than it is.

Start with the Data Profiler to see each column type, the blanks and the duplicate rows. Then use Data Transformation to trim spaces, change date formats, change text case, remove duplicate rows and remove empty values. Use Fill empty cells only when you have a reason, because a filled value changes the data the model sees.

Decide on blanks as a modelling choice, not as a cleaning chore. Sometimes a blank is information. Split your data before you do anything that learns from it, such as scaling or filling with a mean, and do the learning only on the training part. This tool does the structural cleanup. Modelling choices belong in your notebook.

Keep a record. Data Lineage lists the steps applied to the file, so you can reproduce the dataset later. Save the cleaned file as CSV or Parquet, and keep the raw file unchanged.

Open the Data Profiler

Related tool pages

  • CSV Data Quality Checker: Check a CSV for missing values, duplicates, outliers and wrong types in seconds, and share a report. Free, in your browser.
  • Check a CSV for Duplicates and Blanks: Find duplicate rows, empty cells and odd values in a CSV before you use it. See a report per column. Free, in your browser.
  • CSV Data Profiler: Profile a CSV in seconds: column types, missing values, duplicates, outliers and distributions, with a shareable report. Free, in your browser.
  • Open a Large CSV File Online: Open and explore a CSV too big for Excel: see rows and columns, filter with SQL and export. Works for 500 MB and more, in your browser.
  • CSV Viewer Online: Open a CSV file online and see its rows, column types, blanks and duplicates at a glance. Free, private, no upload.
  • Data Profiler Online: Profile a CSV, Excel, JSON or Parquet file online: rows, types, missing values, duplicates and statistics per column. Free, private, large files too.

How to use the Data Profiler: step by step

  1. Profile the file in the Data Profiler.
  2. Open Data Transformation and apply trim, date, case and duplicate steps.
  3. Handle blanks deliberately.
  4. Save as CSV or Parquet and keep the raw file.

When to use the Data Profiler

  • Remove duplicates before a train and test split
  • Fix date and text columns
  • Find columns with too many blanks
  • Keep a record of how a dataset was prepared

Example

Input: id,name,age 1,Ada,36 2,Alan,41

Output: age: integer, 0% missing, mean 38.5, min 36, max 41

Frequently asked questions

What counts as a duplicate row?

A row identical to an earlier one in every column.

What does the Data Profiler show?

Column types, missing values, duplicates, outliers and basic statistics, so you can judge the quality of a file at a glance. You can download the report, with an optional password.

Tip: Remove duplicates before you split into train and test sets, so the same row cannot be in both.