Data, Analytics and AI
Machine learning
Machine learning replaces rules a person writes with rules a program infers from examples, then asks the awkward question of whether they hold on cases it has never met. The ground below runs from supervised and unsupervised learning through the evaluation habits that keep a score honest to where deep learning fits, where it is overkill, and what the entry-level AI exams expect a candidate to know.
Why this topic exists: Supervised and unsupervised learning, model evaluation, deep learning and the discipline of not fooling yourself (SFIA Machine learning; AIF-C01; Azure AI Fundamentals).
Machine learning is inferring a rule from data and judging it by how it performs on examples it did not train on. That second clause is the whole discipline. Anyone can fit a model that describes the past; the work lies in showing it will describe next month, and in noticing when it stops doing so. SFIA frames its machine learning skill around systems that get better with data and cope as conditions shift, which puts monitoring inside the definition rather than after it.
Three families of problem
| Family | The question it answers | How you know it worked |
|---|---|---|
| Supervised learning | Given labelled examples, predict the label of a new case: will this customer cancel, what will this flat sell for | Accuracy, precision and recall, or error, on a held-out test set the model never saw |
| Unsupervised learning | With no labels, find structure: groups of similar customers, unusual transactions, fewer dimensions | Harder to say; usually whether the groups are stable and useful to the people who act on them |
| Reinforcement learning | Choose actions that earn the most reward over time: game play, robot control, some recommendation systems | Reward in a simulator first, then tightly limited trials in the real setting |
Deep learning is a way of tackling any of the three with many-layered neural networks, and it dominates wherever the raw input is images, sound or text. On ordinary business tables of customers and transactions, gradient-boosted trees usually match it for a fraction of the compute and with far easier explanations. Knowing which kind of data you hold is therefore the first design decision, before any choice of library. Transformers, the architecture behind current language models, began life as a deep-learning answer to machine translation.
Not fooling yourself
The beginner's mistake is a model that is too good. Ninety-nine per cent accuracy on a fraud problem where one transaction in a hundred is fraudulent means the model has learned to say no to everything. A near-perfect churn score often means a feature leaked the answer, such as a cancellation date that only exists once the customer has left. The defences are unglamorous: keep a test set you look at once, split by time whenever the future is what you are predicting, compare every model with a naive baseline, and pick a metric that reflects what a wrong answer costs. Record every experiment, the failures included, so that the model you ship can be traced back to the data and settings that produced it.
The fundamentals papers test vocabulary rather than practice. AWS's AI Practitioner and Microsoft's Azure AI Fundamentals ask you to tell learning types apart, recognise evaluation metrics and place services; IBM's free AI Fundamentals covers how machines learn without tying it to a vendor. Google's Advanced Data Analytics certificate comes closest to real practice, with regression and first models in Python notebooks. From here, statistics and experimentation strengthens the foundation, and MLOps and AI engineering covers what happens once a model works.
Next to this topic
- Data analysis fundamentalsAsk, prepare, process, analyse, share: the analyst's workflow as Google's programme and DP-900's core data concepts teach it, before any tool (SFIA Data analytics).
- SQL for analysisQuerying, joining, aggregating and windowing data is the analyst's first real tool and the language of every warehouse (DP-900 relational concepts; DMBOK Data Storage and Operations; the programming field owns database design).
- Data visualisation and business intelligenceDashboards, reports and the semantic models behind them: PL-300's four skill areas and DMBOK's Data Warehousing and BI area (Tableau, Power BI, Looker).
- Statistics and experimentationProbability, inference, regression and A/B tests separate a chart from a claim (Google's Advanced Data Analytics; SFIA Data science).
- Python for data workpandas, NumPy, notebooks and plotting: the working environment of analysis and machine learning (SFIA Data science; the AI pillar's argument for Python).
- Data engineering and pipelinesIngestion, transformation, warehouses, lakehouses and orchestration (DMBOK Data Integration and Interoperability; DEA-C01; DP-700; Databricks; Spark, Airflow, dbt).
- Data governance, quality and privacyOwnership, definitions, metadata, master data, quality rules and lawful use are DMBOK's centre and half its wheel; nobody hires for them until the numbers disagree.
- Generative AI, LLMs and agentsFoundation models, prompting, retrieval, fine-tuning and agents: the newest layer of the field and the one AIF-C01 and Azure AI Fundamentals now test (the AI portal reads the concepts in order).
- Natural language processing and computer visionText and images are the two applied fields where classic machine learning met deep learning first, and where most production AI still lives (SFIA Machine learning; Azure AI Fundamentals AI-901 workloads).
- Responsible AI and AI governanceBias, transparency, privacy, safety and the frameworks that make them work (NIST AI RMF; AIF-C01's responsible-AI domain; SFIA 9's AI and data ethics skill; the EU AI Act).
- MLOps and AI engineeringGetting a model or an LLM application into production and keeping it there: deployment, monitoring, evaluation, cost (SFIA Machine learning and Data engineering; Applied Skills in Foundry; MLOps in the AI portal).
Concepts to know
Glossary entries with the reason each one matters here.
- ML
The topic's own concept.
- Deep Learning
The subfield that took over vision and language.
- Neural nets
The model family behind deep learning.
- Python
The working language.
Certifications that test it
Vendor exams and free certificates; facts, cost and the preparation path are on each page, and the certifications hub has them all.
- AWS · AIF-C01AWS Certified AI PractitionerAWS's AI practitioner exam.
- Microsoft · AI-901Microsoft Certified: Azure AI FundamentalsMicrosoft's AI fundamentals.
- IBM · Course completion · freeAI Fundamentals: Foundations for Understanding AIIBM's free introduction.
- Oracle · 1Z0-1122-26 · freeOracle Cloud Infrastructure AI Foundations AssociateOracle's free AI foundations exam.
Tools of the trade
- Python
Every major ML library is a Python library.
- TensorFlow and Keras
Google's deep-learning framework.
- PyTorch
The research community's deep-learning framework.
Frequently asked
- Should I learn statistics before machine learning?
- Enough to understand sampling, variance and why a test set exists. Without that, machine learning becomes trial and error against a leaderboard. Deeper theory can come later, but the reflex of asking whether a result would survive fresh data should come first.
- Do I need a GPU to learn machine learning?
- No. Classic methods on tabular data run comfortably on a laptop, and that is where most learning happens. Deep learning on images or text benefits from a GPU, which free notebook services offer in limited amounts. Buy hardware only once you know what you will train on it.
- What is overfitting?
- A model that has memorised the quirks of its training data instead of the pattern behind them. It scores brilliantly on examples it has seen and poorly on new ones. Held-out test data, simpler models, regularisation and more varied training data are the usual remedies.
- scikit-learn, TensorFlow or PyTorch?
- scikit-learn first, for classic methods and the habits of splitting, pipelines and evaluation it builds in. For deep learning, PyTorch is where most research and a large share of industry work now happens, while TensorFlow remains common in older production systems. The ideas carry across and the syntax takes days.
Courses in the directory
1,018 courses are filed here; the top 6 by our ranking, details and the provider link on each course page.
Last reviewed 26 September 2026 · Getting Digital