Tabular Data

Tabular data is data organised in a table — rows and columns, like a spreadsheet. Each row is a record (a customer, a transaction, an event), and each column is a feature (name, date, amount, category). It is the most common data format in business and the foundation of most analytics and traditional machine learning.

Why tabular data is important

Despite the excitement around AI processing images, text, and video, the vast majority of business decisions still rely on tabular data. Your CRM, ERP, financial systems, HR database, and operational dashboards all store tabular data. If you want to predict customer churn, forecast revenue, or optimise pricing, you are working with tables.

AI and tabular data

Different AI approaches handle tabular data differently:

• Traditional ML (random forests, gradient boosting): These algorithms were designed for tabular data and remain the best performers for many structured data tasks. XGBoost and LightGBM consistently win machine learning competitions on tabular datasets.
• Deep learning: Neural networks can process tabular data but typically do not outperform traditional ML on structured tables. Deep learning shines on unstructured data (images, text, audio).
• LLMs: You can paste a table into Claude or ChatGPT for ad-hoc analysis. This is convenient for quick insights but not suitable for production-scale analytics.

Common operations

Working with tabular data involves standard operations:

• Filtering: Show only rows that meet certain criteria
• Sorting: Order rows by a specific column
• Aggregation: Calculate totals, averages, or counts by group
• Joining: Combine tables that share a common key
• Pivoting: Reorganise data to compare values across categories

Data quality challenges

Tabular data in the real world is rarely clean:

• Missing values in critical columns
• Inconsistent formatting (different date formats, varied spellings)
• Outliers that skew analysis
• Duplicate records from system integrations
• Columns that contain mixed data types

Preparing tabular data for AI

Before training a machine learning model on tabular data, you typically need to:

• Handle missing values (fill, remove, or flag them)
• Encode categorical variables (convert "red," "blue," "green" to numbers)
• Normalise numeric features (scale to a common range)
• Split into training, validation, and test sets