FROM PRACTICE / INTO INTERACTION
AI & Data Jargon Buster.
Explore 123 terms in plain English and technical language. Search, filter, compare and connect an idea to a working example. Some ideas arrive wearing rather more syllables than they need.
Loading the interactive studio…
Read more · Searchable reference for all 123 terms
Big data
Very large, fast, or messy data that normal spreadsheets can’t handle well.
Datasets whose size, speed, or complexity require scalable storage/compute (often distributed systems).
3Vs (Volume, Velocity, Variety)
How much data; how fast it comes in; and how different/ messy it is.
Common framing of big data challenges: amount, arrival speed, and data-type diversity.
Data warehouse
A tidy reporting store for dashboards and KPI packs.
Structured, curated repository optimized for analytics and reporting (schema-on-write).
Data lake
A big storage bucket for all data types before cleaning/curation.
Centralized storage for raw/semi-processed structured & unstructured data (schema-on-read).
Cloud computing
Renting computing power and storage instead of buying servers.
On-demand compute/storage services delivered over the internet (IaaS/PaaS/SaaS).
Edge computing
Doing some computing locally so you don’t have to send everything to the cloud.
Processing data near the data source rather than in a central cloud/data center.
Batch processing
Run the job nightly/weekly rather than live.
Processing data in groups at scheduled intervals.
Stream processing
Live updating as data comes in.
Continuous processing of events/records as they arrive in near real time.
Structured / semi-structured / unstructured data
Tables vs. ‘sort-of tables’ vs. emails/PDFs/notes.
Structured: fixed schema; semi: flexible schema (JSON); unstructured: free text/media.
Data governance
Rules and ownership for keeping data safe and trustworthy.
Processes, roles, policies, and controls that ensure data quality, security, and compliance.
Privacy / consent
People should know what’s collected and why; only use it appropriately.
Legal/ethical requirements for collecting and using personal data with appropriate lawful basis and transparency.
Proxy data
A ‘best guess’ indicator for something you can’t measure directly.
A variable used as an indirect stand-in for another harder-to-measure variable.
Anomaly detection
Spotting ‘odd’ cases worth investigating.
Methods to identify rare/unusual patterns relative to expected behavior.
Text analytics / NLP
Getting insights from words at scale.
Computational methods for extracting meaning from text (tokenization, embeddings, classification).
Sentiment analysis
A model suggests a tone label for words; it does not measure a person’s feelings.
Classifying expressed sentiment in text under a defined labelling scheme.
Artificial intelligence (AI)
Software doing ‘smart’ tasks like understanding text or making predictions.
Systems that perform tasks normally requiring human intelligence (perception, reasoning, language).
Machine learning (ML)
Models that learn from examples rather than being hand-coded.
Algorithms that learn patterns from data to make predictions/decisions without explicit rules.
Generative AI
AI that can write, summarise, or draft code.
Models that generate new content (text/code/images) by learning from large datasets.
Supervised learning
Learning from past cases where we know the outcome.
Learning from labelled examples to predict a target (classification/regression).
Unsupervised learning
Finding groups/patterns when we don’t have ‘right answers’.
Finding structure in unlabelled data (clustering, dimensionality reduction).
Reinforcement learning
Learning by trial and error with points for good outcomes.
Learning a policy via reward feedback in an environment (exploration vs exploitation).
Feature
A data point the model uses to decide.
Input variable used by a model; may be raw or engineered.
Feature engineering
Turning messy fields into useful signals.
Transforming raw data into informative features (encoding, aggregations, interactions).
Encoding
Turning words/categories into numbers a model can use.
Converting categorical variables to numeric representations (one-hot, ordinal, target encoding).
Model evaluation
Checking if the model works reliably, not just on training data.
Measuring generalization performance with metrics and validation procedures.
Bias (ML)
A model can make systematic errors or reproduce unfair patterns. These are related but distinct meanings of bias.
Bias can mean systematic estimation error or assumptions in a model; social bias concerns harmful patterns in data, use and outcomes.
scikit-learn
A toolkit for building common ML models quickly in Python.
Python library offering classical ML algorithms, preprocessing, and evaluation tools.
PyTorch
A powerful library for building neural nets.
Deep learning framework using dynamic computation graphs for neural networks.
Keras
Simpler interface for neural network models.
High-level neural network API (often on TensorFlow) for quick model building.
XGBoost
A strong ‘go-to’ predictor for structured/tabular data.
Gradient-boosted decision tree implementation optimized for performance.
Hugging Face
A library and hub of ready-made AI language models.
Ecosystem for pretrained NLP/vision models, datasets, and tooling (Transformers).
Data science
Using data + methods to answer questions and support decisions.
Interdisciplinary practice of extracting insight/value from data using stats, programming, and domain knowledge.
CRISP-DM
A repeatable project lifecycle for data work.
Business understanding → data understanding → preparation → modelling → evaluation → deployment.
Business intelligence (BI)
Dashboards and reports showing what happened.
Reporting/visualisation focused on descriptive analytics and decision support.
Descriptive / Predictive / Prescriptive analytics
What happened; what might happen; which action may help under stated assumptions.
Descriptive analysis summarises observations; predictive analysis estimates unobserved outcomes; prescriptive methods compare actions under objectives and constraints.
Data labelling
Give examples a reviewable label; a label is not automatically objective truth.
Assigning annotations according to a documented scheme, often with ambiguity and disagreement.
Sampling
Looking at a smaller slice to learn about the whole.
Selecting a subset of data to estimate properties of a population.
Bayes' theorem
A way to revise your belief when you get new info.
Updates probability based on evidence: P(A|B) ∝ P(B|A)P(A).
Interpretability
Can we explain the model’s decisions?
Degree to which a human can understand model behavior and reasons for predictions.
Explainable AI (XAI)
Tools that show what drove a prediction.
Techniques that provide explanations of model outputs (feature importance, SHAP/LIME).
Dimensionality reduction
Compressing many columns into fewer, without losing too much info.
Techniques to reduce feature space while preserving signal (PCA, t-SNE, UMAP).
Clustering
Automatically grouping similar things.
Grouping similar data points without labels (k-means, DBSCAN, hierarchical).
Descriptive statistics
Quick numbers that describe the data at a glance.
Summary measures of a dataset (mean, median, variance, percentiles).
Inferential statistics
Using a sample to draw conclusions about the whole group.
Methods to infer population properties from a sample (tests, confidence intervals).
Hypothesis testing
Test evidence against an assumption; a small p-value does not prove an effect is important or causal.
Assessing how compatible observed data are with a specified null model under stated assumptions.
Type I error
Thinking there’s a real change when there isn’t.
False positive: rejecting a true null hypothesis.
Type II error
Missing a real improvement/problem.
False negative: failing to reject a false null hypothesis.
Correlation vs causation
Two things moving together doesn’t mean one causes the other.
Correlation measures association; causation requires evidence of causal mechanism and controls.
ANOVA
Checks whether averages differ between multiple groups.
Analysis of variance; tests differences in means across groups.
Chi-squared test
Checks whether categories are related.
Tests association between categorical variables.
Linear regression
A line (or plane) that predicts a number.
Models a continuous outcome as a linear function of features; estimated via least squares.
Prompt engineering
Asking AI the right way to get better results.
Designing prompts to elicit reliable, useful outputs from LLMs.
Hallucination (LLMs)
AI confidently making things up.
Model-generated output that is plausible but incorrect or unsupported.
Variable
A labelled box that stores a value.
Named reference to a value in memory (may be mutable depending on type).
Data types
Different kinds of information (numbers, text, yes/no, lists).
Classification of values (int, float, str, bool, list, dict, tuple).
Function
A mini-machine you can reuse.
Reusable block of code with inputs (parameters) and outputs (return value).
Loop (for/while)
Do something for each item; or until a rule is met.
Control structure that repeats over items or while condition holds.
List / tuple / dictionary
Collections: editable list; fixed list; and lookup table.
Common Python collections: list mutable; tuple immutable; dict key→value map.
Module
A bundle of code you can plug in.
A Python file/package that provides reusable functions/classes (importable).
Object-oriented programming (OOP)
Grouping data + actions together in ‘objects’.
Paradigm organizing code around objects with state and behavior (classes).
DataFrame
A spreadsheet-like table in Python.
2D tabular data structure (rows/columns) in pandas.
Missing values (NaN)
Blank cells that can break analysis if ignored.
Absent data represented as NaN/None; requires imputation or removal strategies.
Outlier
A weird value that doesn’t fit the rest.
Observation that deviates markedly from others; may be error or rare valid event.
Matplotlib
A library to make charts in Python.
Python plotting library for creating static visualizations.
Seaborn
Easier charts for exploring data patterns.
Statistical visualization library built on matplotlib with high-level functions.
Pearson correlation
How strongly two numbers move together in a straight-line way.
Linear correlation coefficient between two continuous variables (-1 to 1).
Spearman correlation
Correlation based on ordering rather than exact values.
Rank-based correlation capturing monotonic relationships; robust to nonlinearity/outliers.
Web scraping
Automatically collecting data from websites.
Programmatic extraction of data from webpages using HTTP requests and parsers.
BeautifulSoup
Tool to read webpage code and pick out what you need.
Python library for parsing HTML/XML documents for scraping.
Hyperparameter tuning
Trying different model settings to get better results.
Searching parameter settings to improve model performance (grid/random/Bayesian search).
Streamlit
Turn a notebook into a simple web dashboard.
Framework to build data apps and dashboards in Python with minimal code.
Train/validation/test split
Splitting data so we learn, tune, then test fairly.
Partitioning data into sets for training, tuning, and unbiased evaluation.
Cross-validation (k-fold)
Testing the model multiple times on different slices.
Repeatedly training/testing across folds to estimate generalization reliably.
Data leakage
A model gains information it should not have, including test-set information or future outcomes.
Information unavailable at prediction time or reserved for evaluation improperly influencing model fitting or selection.
Baseline model
A minimal “starting point” to beat.
Simple reference model (e.g., majority class, mean) used for comparison.
Classification
Predicting a category like Yes/No.
Task of predicting discrete categories/labels.
Multiclass classification
Predicting one of many categories.
Classification with more than two classes.
Time series
Data tracked over time.
Ordered data points over time, often autocorrelated and seasonal.
Seasonality
Regular ups and downs that repeat.
Recurring patterns at regular intervals (weekly/termly/annual).
Concept drift
The world changes and the model goes stale.
When data patterns change over time so model performance degrades.
Model monitoring
Keeping an eye on a model after go-live.
Tracking model performance and data characteristics in production.
Precision
When we flag something, how often are we right?
TP/(TP+FP): proportion of predicted positives that are correct.
Recall (Sensitivity)
How many real problems we successfully catch.
TP/(TP+FN): proportion of actual positives caught.
F1 score
Balances “right flags” and “misses.”
Harmonic mean of precision and recall.
Class imbalance
When “Yes” cases are rare compared to “No.”
When one class is much rarer than another, skewing training/metrics.
SMOTE
Creating extra “rare” examples to help learning.
Synthetic Minority Over-sampling Technique; creates synthetic samples for minority class.
ROC curve
Shows trade-offs as we change the “flag” threshold.
Plot of TPR vs FPR across thresholds for binary classifiers.
Calibration
Does “70% risk” really mean ~70% happen?
Agreement between predicted probabilities and observed frequencies.
Brier score
How good our probability forecasts are.
Mean squared error of probabilistic predictions vs outcomes.
Threshold
The cut-off for “flag vs not flag.”
Value used to convert probabilities/scores into class decisions.
Confounding variable
A shared cause can explain an apparent relationship; it need not be hidden.
A variable that influences the exposure and outcome and can distort a causal comparison.
A/B testing
Testing two options to see which works better.
Randomized experiment comparing two variants to measure impact.
Causal inference
Figuring out what actually caused the change.
Methods for estimating cause-effect relationships from data/experiments.
Counterfactual
What would have happened otherwise.
Hypothetical outcome under an alternative scenario (“what if”).
Uplift modelling
Who benefits most from an action.
Predicts incremental impact of an intervention on an outcome.
Decision tree
A flowchart-like predictor.
Tree-structured model splitting on features to predict outcomes.
Random forest
Many trees voting together.
Ensemble of decision trees trained on bootstrapped samples with feature randomness.
Gradient descent
How many models “learn” by gradually reducing error.
Optimization method that iteratively reduces loss by stepping along gradients.
Loss function
The score the model tries to improve.
Objective function minimized during training (e.g., MSE, log loss).
Learning rate
How big each learning step is.
Step size in gradient-based optimization.
Regularization (L1/L2)
Discouraging overly complex models.
Penalties added to loss to reduce overfitting (L1 sparsity, L2 shrinkage).
Lasso / Ridge regression
Penalties can reduce overfitting; they do not guarantee a model will generalise.
Linear models penalising coefficients with L1 or L2 penalties to control complexity; tuning and validation remain necessary.
Principal Component Analysis (PCA)
Compressing many columns into a few summary dimensions.
Linear dimensionality reduction via orthogonal components maximizing variance.
Embeddings
Numbers that represent meaning so machines can compare text.
Vector representations of text/items capturing semantic similarity.
Vector database
A search store for “meaning-based” lookup.
Database optimized for similarity search over embedding vectors.
Cosine similarity
Compare vector directions. A high value is not proof that two texts mean the same thing.
The normalised dot product of two nonzero vectors; interpretation depends on their representation.
Retrieval-Augmented Generation (RAG)
Give an AI relevant source material before it answers; retrieval does not guarantee a correct or faithful answer.
A system retrieves external material and supplies it to a generative model as context.
Prompt injection
Tricking the AI to ignore rules or reveal data.
Attacks that manipulate prompts/instructions to override safeguards or exfiltrate data.
PII (Personally Identifiable Information)
Information that points to a person.
Data that can identify an individual (directly or indirectly).
Anonymization vs pseudonymization
Removing names is not enough to guarantee anonymity. Coded records can still relate to identifiable people.
Anonymisation aims to prevent identification in context; pseudonymisation replaces identifiers while linkage or re-identification may remain possible.
Differential privacy
A carefully calibrated method limits what outputs reveal about participation; adding arbitrary noise is not enough.
A formal guarantee bounding how much an output distribution changes when one person’s contribution changes, governed by explicit parameters and composition.
Data lineage
Knowing where data came from and what happened to it.
Tracing data’s origin, transformations, and movement across systems.
Metadata
Labels and context for understanding data.
Data about data (definitions, ownership, timing, quality, provenance).
Data dictionary
A glossary for dataset columns.
Structured documentation of fields: meaning, type, allowed values, owner.
ETL / ELT
Moving and reshaping data for use.
Extract-Transform-Load vs Extract-Load-Transform data pipelines.
Orchestration (Airflow)
Running data jobs in the right order on a schedule.
Coordinating scheduled data workflows with dependencies and monitoring.
Reproducibility
Getting the same answer again reliably.
Ability to recreate results with same code/data/config versions.
Version control (Git)
“Track changes” for code.
System for tracking changes to code and collaborating (commits/branches).
Experiment tracking (MLflow)
A record of model runs and what changed.
Logging params, metrics, artifacts to compare ML runs systematically.
Model registry
Keep track of model versions and their status; registration does not itself mean a model is approved.
A versioned catalogue of models, metadata and lifecycle states, supporting review and controlled promotion.
Model card
A one-page “nutrition label” for a model.
Standard documentation describing model purpose, data, metrics, limitations, ethics.
Human-in-the-loop (HITL)
A person has a review role; that only helps when they have enough information, time and authority to intervene.
A workflow gives people a specified role in reviewing or influencing system outputs or actions.
Automation bias
People follow the AI even if it’s mistaken.
Humans over-trust automated suggestions even when wrong.
EXTERNAL RESOURCES / GLOBAL BA STANDARDS
Business analysis, grounded in shared practice.
Use these references for the wider professional context behind the glossary. The definitions here are teaching explanations, not official IIBA definitions.
IIBA · BABOK® Guide, Version 3 — a globally recognized professional standard for business analysis, covering knowledge areas, tasks, competencies and techniques.
Read more · Official resources and how to use them
IIBA · The Business Analysis Standard offers a concise foundation and a free PDF through IIBA. It is a separate, updated resource—not a new numbered edition of BABOK.
BABOK v3 corrections and errata · IIBA KnowledgeHub
Edition check: 9 October 2026. IIBA’s current resources continue to identify BABOK v3. Access the full guide through IIBA’s membership or purchase options. These links support independent learning; they do not imply IIBA endorsement or certification of this site or its tools.
Personal tool suggestions
Enable JavaScript to choose optional feedback-based suggestions.
EXPLAINABLE AI & DATA / READ THE METHOD
A field guide to AI, data and decisions.
Capabilities, algorithms, packages and delivery stages are different things. This reference defines them and points to relevant examples; it does not claim that every method runs here.
Follow the reasoning.
- CapabilityWhat kind of task?
- AlgorithmWhat mathematical method?
- PackageWhich implementation toolbox?
- PipelineHow is it checked and used?
Explore the building blocks.
Number of concepts in this guide, not performance or usage measures.
Explore the explanation matrix · 45 concepts
45 concepts shown
On small screens, scroll the table sideways. “Learning context” explains a related idea; it does not mean that method runs on this page. Deep learning is the standard term for “deep ML”.
| Technical term | Technical description / definition | In plain English | How this page uses it |
|---|---|---|---|
| Machine learning (ML)AI Capability | Methods that estimate patterns or parameters from examples to support inference on other inputs.Primary reference ↗ | Learn a pattern from examples, then check whether it helps with new ones. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Deep learningAI Capability | Machine learning using neural networks with multiple representation-learning layers.Primary reference ↗ | A layered model learns useful representations; more layers do not guarantee a better answer. | Learning reference Deep learning is included here as general learning context, not a claim about a private project. |
| Natural language processing (NLP)AI Capability | Computational methods for representing, analysing or generating human language.Primary reference ↗ | Help a computer work with words, while remembering that context matters. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Sentiment analysisAI Capability | A language-analysis task that estimates an expressed attitude, often as polarity or sentiment categories.Primary reference ↗ | Estimate the tone of words—not the private feelings or personality of their author. | Learning reference A language-analysis capability to distinguish from the Applied Analytics topic classifier; no teammate sentiment profiling. |
| Computer visionAI Capability | Computational analysis of image or video content for tasks such as representation, detection or classification.Primary reference ↗ | Turn visual patterns into useful signals; the image still needs context. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Large language model (LLM)AI Capability | A language model trained on substantial data to estimate and generate token sequences from context.Primary reference ↗ | A fluent drafting partner whose statements still need checking. | Learning reference Private PW and Ilé: explicit reviewed Send invokes the selected language-model service. |
| Descriptive analyticsAI/Data Concepts | Summaries of observed data using counts, rates, distributions or comparisons.Primary reference ↗ | Tell us what the records show; this alone does not predict next week. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Features and labelsAI/Data Concepts | Features are model inputs; labels are target values or categories used in supervised learning.Primary reference ↗ | What the model sees, and the answer it is learning to estimate. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Synthetic dataAI/Data Concepts | Generated records used to represent scenarios rather than directly report observed individuals or events.Primary reference ↗ | A practice dataset. Useful for demonstration, not proof of workplace impact. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Probability and uncertaintyAI/Data Concepts | A probability quantifies an event under a model and assumptions; it is not a guarantee.Primary reference ↗ | A chance, not a promise. Ask what evidence and assumptions produced it. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Rates and denominatorsAI/Data Concepts | A rate or share relates a count to a defined exposure or eligible population.Primary reference ↗ | Keep the number underneath the fraction visible when comparing people or periods. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Association is not causationAI/Data Concepts | An observed relationship does not alone identify the effect of changing one variable.Primary reference ↗ | Two things moving together does not prove that one caused the other. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Random forestAlgorithm | An ensemble of decision trees whose outputs are combined for classification or regression.Primary reference ↗ | Several branching rulebooks contribute to one estimate. | Learning reference Procurement Lab: synthetic delay classification; Applied Analytics: exported classifier. |
| Ridge regressionAlgorithm | A linear regression model with an L2 coefficient penalty to limit excessive parameter magnitude.Primary reference ↗ | Fit a trend while discouraging overly extreme weights. | Learning reference Procurement and Applied Analytics labs: exported linear numerical predictions. |
| Logistic regressionAlgorithm | A linear predictor mapped through a logistic function to estimate a binary-class probability.Primary reference ↗ | Combine weighted clues into a chance of one of two outcomes. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Multinomial Naive BayesAlgorithm | A probabilistic classifier using feature counts and class-conditional independence assumptions.Primary reference ↗ | Count familiar words and compare which labelled themes they tend to accompany. | Learning reference Applied Analytics: vocabulary-based text theme classification. |
| Gradient boostingAlgorithm | An additive ensemble trained sequentially to improve a specified loss objective.Primary reference ↗ | Build a sequence of small models that work on what earlier ones missed. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Isolation forestAlgorithm | An anomaly detector based on how quickly random partition trees isolate an observation.Primary reference ↗ | Flag unusual records for review; unusual does not mean fraudulent or wrong. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| K-means clusteringAlgorithm | Partitions observations into k groups by iteratively assigning points to centres and updating those centres.Primary reference ↗ | Group similar patterns. Group labels are not rankings or explanations of cause. | Learning reference Chelsfield: standardised team-colour statistical clusters, prepared in Python. |
| Rule-based logicAlgorithm | Explicit authored conditions map inputs to outputs without fitting parameters from training examples.Primary reference ↗ | A transparent set of if-this-then-that instructions. | Learning reference BA workbenches and cultural learning activities: authored choices and deterministic outputs. |
| Jaccard similarityAlgorithm | The size of the intersection of two sets divided by the size of their union.Primary reference ↗ | How much two collections of tags overlap. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Cosine similarityAlgorithm | The normalised dot product of two nonzero feature vectors.Primary reference ↗ | Compare the direction of two representations, rather than simply their size. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| EmbeddingAI/Data Concepts | A numerical vector representation in which distances or similarity can encode learned relationships.Primary reference ↗ | A numerical description for comparing content—not a copy of its meaning. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Data science pipelineData Science Pipeline | An ordered, reproducible sequence for acquiring, validating, preparing, analysing and evaluating data.Primary reference ↗ | A chain from a question and source to an answer someone can check. | Learning reference Chelsfield: public-source import, validation, historical calculation, frozen forecasts and subsequent scoring. |
| Extraction, transformation and provenanceData Science Pipeline | ETL moves data from sources through defined transformations; provenance records its origin and changes.Primary reference ↗ | Read it, tidy it, and keep the receipt. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Data quality checksData Science Pipeline | Validation of completeness, allowed values, uniqueness, types and consistency with expected constraints.Primary reference ↗ | Check missing, repeated and impossible entries before trusting a chart. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Retrieval and groundingData Science Pipeline | Selecting relevant source material to supply context for an answer, with traceable references.Primary reference ↗ | Bring the relevant pages to the conversation; do not pretend every page was read. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Held-out evaluationEvaluation & XAI | Assessment on examples excluded from fitting and model selection, according to the declared split.Primary reference ↗ | Keep some answers out of practice so the test means something. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Confusion matrix, precision and recallEvaluation & XAI | A confusion matrix counts predicted versus actual classes. Precision measures positive-result reliability; recall measures positive-case coverage.Primary reference ↗ | Which alarms were right, which were false, and which cases were missed? | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Residuals and prediction errorEvaluation & XAI | A residual compares an observed value with its fitted or predicted value. Error metrics summarise those discrepancies.Primary reference ↗ | How far did the estimate miss, and where does it miss most? | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Explanation versus causal proofEvaluation & XAI | A model explanation describes selected behaviour or contributions; it does not establish causal effects or full reliability.Primary reference ↗ | A reason for this output is useful, but it is not proof that the model is right. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Human oversight and consentResponsible AI | People retain appropriate review authority, permission controls and correction routes for data use and outputs.Primary reference ↗ | Keep someone able to question, correct or stop the result. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Data minimisationResponsible AI | Limit collected, disclosed and retained information to what is needed for the stated purpose.Primary reference ↗ | Use the smallest appropriate amount of information. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Coverage, bias and representationResponsible AI | Assess how sampling, labels, missing groups and design choices affect error and applicability.Primary reference ↗ | Ask whose experience the data includes—and whose it misses. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| NumPyPython Package | A Python library for multidimensional arrays and numerical operations.Primary reference ↗ | The numerical toolbox behind many Python calculations. | Learning reference Used during Python model/data preparation and football numerical analysis. |
| pandasPython Package | A Python library for labelled tabular data manipulation and analysis.Primary reference ↗ | A programmable workbench for cleaning and comparing tables. | Learning reference A useful tabular Python package presented for learning; not claimed as a dependency of every site tool. |
| scikit-learnPython Package | A Python library providing estimators, preprocessing tools and model-evaluation utilities.Primary reference ↗ | A toolkit of models and fairer ways to test them. | Learning reference Used by the lab training scripts and Chelsfield clustering; the public browser does not install Python packages. |
| MatplotlibPython Package | A Python plotting library for static, animated and interactive visualisations.Primary reference ↗ | Turn numerical work into labelled figures that can be inspected and shared. | Learning reference A Python visualisation package presented for learning; charts here may also be exported SVGs or JavaScript. |
| Python standard libraryPython Package | Modules distributed with Python for tasks including parsing, file handling, dates and structured data.Primary reference ↗ | The tools already in Python’s kit; not a separate AI model. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Visual encodingVisualisation | Mapping values or categories to marks, position, length, colour and labels.Primary reference ↗ | Make the comparison visible, with a scale and a clear label. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Geospatial layers and projectionVisualisation | Coordinates are transformed to a map surface; separate layers encode geographic features or interpretations.Primary reference ↗ | A map combines evidence layers. Its boundary lines need a date and a meaning. | Learning reference Yorùbáland atlas: geographic layers, comparisons and interpretation notes. |
| Unicode normalisationAlgorithm | Transforming equivalent character sequences into a consistent Unicode representation.Primary reference ↗ | Help differently encoded letters match without throwing away meaningful accents. | Learning reference Cultural AI language experiment: representation and preservation of marks. |
| Word error rate (WER)Evaluation & XAI | Substitutions, deletions and insertions divided by the reference word count.Primary reference ↗ | Compare a transcript against a checked reference; the number alone cannot judge cultural meaning. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Decision thresholdEvaluation & XAI | A cutoff that converts a model score into an action or predicted category.Primary reference ↗ | Decide how strong a signal must be before you act; changing it changes mistakes. | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
| Ranking evaluationEvaluation & XAI | Measures such as Precision@k and NDCG@k evaluate relevance among ordered recommendations.Primary reference ↗ | Were the most helpful suggestions near the top? | Learning reference Use the linked definition and context panels on individual pages to distinguish a learned model, a rule-based tool and published evidence. |
No matching concepts. Choose another category or clear the search.
