Data cleaning is the process of detecting and correcting errors, inconsistencies and irrelevant records in raw data so that it can be used effectively for analysis, statistics and machine learning. It is a foundational step of data preprocessing.
Raw data is often noisy, incomplete and inconsistent, and these flaws directly reduce the accuracy of a model. Clean data also matters for EDA (Exploratory Data Analysis): charts and summaries are only trustworthy when the underlying data is correct, which helps people make the right decisions from insights.
What cleaning covers
- Finding and removing duplicate records
- Fixing wrong data types and inconsistent formats
- Correcting spelling and naming inconsistencies
- Handling missing values
- Detecting and treating outliers
- Dropping irrelevant variables
Why do errors appear in data?
- Human error — typing mistakes, skipped fields.
- System failures — sensor faults, broken data transfers.
- Integration problems — merging data from different sources with different formats or units.
- Collection issues — poorly designed forms, changing definitions over time.
Cleaning is iterative
Cleaning is rarely a one-shot job. The usual loop is: inspect → fix → verify → repeat until the data quality checks pass.