Machine Learning and Data Terms: A Glossary

Part of the Five Stones knowledge base - 38 terms on machine learning and data, in plain English.

Machine Learning (ML) - A branch of AI where a system improves at a task by learning from data, rather than being explicitly programmed with fixed rules.
Supervised learning - Training an AI model on data that is already labelled with the correct answer, so it learns to predict that answer on new data.
Unsupervised learning - Training an AI model on unlabelled data so it finds patterns or groupings on its own, without being told the "right" answer.
Reinforcement learning - Training an AI by rewarding good outcomes and penalising bad ones, so it learns through trial and error.
Predictive analytics - Using historical data and AI/statistics to forecast a future outcome, such as which customers are likely to churn.
Prescriptive analytics - Going a step beyond prediction to recommend a specific action, such as which customer to call first this week.
Structured data - Information organised in a fixed format, like spreadsheet rows and columns, which is easy for software to search and analyse.
Unstructured data - Information with no fixed format, like emails, voice notes or photos, which needs AI to make it searchable or usable at scale.
Data cleaning - Fixing errors, duplicates and gaps in a dataset before it is used, since messy data produces unreliable AI output.
Data pipeline - The automated process that moves data from where it is created (e.g. a booking form) to where it is used (e.g. a dashboard or AI model).
Data warehouse - A central place where a business's data from different systems is combined and stored for analysis.
Data lake - A large storage pool that holds raw data in its original format, structured or not, until it is needed.
ETL (Extract, Transform, Load) - The standard process of pulling data out of one system, reshaping it, and loading it into another, usually to feed a report or an AI tool.
Database - The underlying system that stores a business's records (customers, orders, appointments) in an organised, searchable way.
CRM data - Customer information (contact details, purchase history, interactions) stored in a Customer Relationship Management system.
Data silo - Information trapped in one system or department, unable to be seen or used by another part of the business, a common blocker to AI adoption.
Data integration - Connecting different systems (e.g. your accounting software and your booking system) so data flows between them automatically.
API integration - Connecting two pieces of software using their APIs so they can share data or trigger actions in each other automatically.
Real-time data - Information updated instantly as events happen, as opposed to being updated in a daily or weekly batch.
Batch processing - Handling a large group of tasks or data together at set intervals (e.g. overnight), rather than one at a time as they arrive.
Data governance - The rules and responsibilities a business sets for who can access, change or delete its data, and how it is kept accurate.
Data quality - How accurate, complete and consistent a business's data is, which directly determines whether AI built on it can be trusted.
Feature (in ML) - An individual measurable input a model uses to make a prediction, such as "number of days since last visit."
Overfitting - When an AI model learns its training data too specifically and performs poorly on new, real-world data as a result.
Model drift - When an AI model's accuracy quietly degrades over time because real-world conditions have changed since it was trained.
Labelling (data labelling) - Tagging raw data with the correct answer (e.g. marking an email as "spam" or "not spam") so a model can learn from it.
Data annotation - Another term for labelling: adding notes or tags to data so it becomes usable for training an AI model.
Dataset - A defined collection of data used to train, test or evaluate an AI model.
Data privacy - The principle and practice of controlling who can access personal or sensitive information a business holds.
Data residency - Where, geographically, a business's data is physically stored, which matters for compliance with local laws.
Data anonymisation - Removing or masking identifying details from a dataset so individuals cannot be recognised from it.
Data breach - An incident where protected data is accessed, leaked or stolen without authorisation.
Business intelligence (BI) - Tools and processes that turn raw business data into dashboards and reports a human can act on.
Dashboard - A visual, at-a-glance screen showing a business's key numbers, usually pulled automatically from underlying systems.
KPI (Key Performance Indicator) - A specific number a business tracks to judge whether it is succeeding at something important, e.g. response time or conversion rate.
Data-driven decision-making - Making business choices based on what the numbers show, rather than instinct alone.
Historical data - Past records used to spot patterns, forecast trends, or train a predictive model.
Real-world data - Data collected from actual operations, as opposed to synthetic or test data.

Looking for a different topic? Browse the full knowledge base, or tell us what's missing.