BTEC Level 3 Unit 21 Introduction to Artificial Intelligence Answer Guide

This BTEC Unit 21 Introduction to Artificial Intelligence (AI) answer guide covers the assignment from the Pearson BTEC International Level 3 qualifications in Information Technology, in which you investigate AI, prepare a data set and train your first machine learning model. The unit is internally assessed through up to two assignments (learning aim A, then aims B and C together). This guide explains what each Pass, Merit and Distinction criterion asks, gives a structure for every task, and includes model paragraphs, a short Python example and a fully worked model-evaluation calculation.

What the Unit 21 AI assignment asks you to do

You act as an independent junior developer building a portfolio. You choose a real-world problem in an area you care about — sport, health, music, retail, the environment — and solve it with AI. Every learner’s project should be different, with its own data, goals and model. The brief sets three tasks:

• Task 1 – AI investigation (Learning aim A): define AI and its core concepts, classify your project as weak (narrow) or strong (general) AI, identify the type and subset of AI it uses, explain the AI pipeline, and discuss the ethical and legal implications.
• Task 2 – Data for my project (Learning aim B): set project objectives, identify data sources, describe the data using the five Vs of big data, and clean the data with evidence of each step.
• Task 3 – Building the AI solution (Learning aim C): choose a suitable model, write it in Python using libraries such as scikit-learn or TensorFlow, split the data 80/20 into training and test sets, report accuracy and other metrics, then refine and evaluate.

Learning aims for Unit 21

Pass, Merit and Distinction criteria explained

The wording above follows the linked centre brief, which also gives the criterion codes; we have not been able to check the codes against the Pearson unit specification, so use the codes printed on your own brief. That brief prints the same wording for the aim B and aim C Distinction criteria. Ask your tutor how they want the two separated; a sensible split is to evaluate your data preparation for aim B and your model and its performance for aim C.

How to answer Learning aim A: investigating AI

1. Define AI and key terms: algorithm, model, training, machine learning, deep learning.
2. Weak (narrow) AI versus strong (general) AI. Almost every real project, including yours, is narrow AI.
3. Types and subsets: supervised, unsupervised and reinforcement learning; natural language processing; computer vision.
4. The AI pipeline: problem definition, data collection, data preparation, model training, evaluation, deployment and monitoring.
5. Industry uses, benefits and risks in at least two sectors.
6. Ethical and legal issues: bias and fairness, transparency, privacy, accountability and job change. Where UK law applies, refer to UK GDPR and the Data Protection Act 2018, the Equality Act 2010 and the Copyright, Designs and Patents Act 1988; as an International learner, also mention your own country’s data protection law.

How to answer Learning aim B: planning and preparing data

Start with two or three SMART objectives, such as “classify match results as win, draw or loss with at least 70% accuracy on unseen data”. Then document your data.

• Source and permission: where the data came from (open data portal, public data set repository, your own survey) and its licence.
• Format: CSV, JSON, images; number of rows and columns; data types.
• Five Vs: volume, velocity, variety, veracity and value — each applied to your data, not just defined.
• Cleaning steps with before/after screenshots: removing duplicates, handling missing values, correcting types, encoding categories, scaling numbers.
• Refinement for the aim B Merit: explain why each change should improve the model, for example removing a column that leaks the answer.

How to answer Learning aim C: building and testing the model

• Justify your model: for example, a decision tree is easy to explain; logistic regression suits two-class problems; k-nearest neighbours is simple but slow on large data.
• Show commented code and output screenshots.
• Use an 80/20 train-test split with a fixed random state so results can be repeated.
• Report accuracy, plus a confusion matrix, precision and recall where classes are uneven.
• Refine (aim C Merit): change one thing at a time — tree depth, features, number of neighbours — and record the result in a table.

A minimal scikit-learn structure looks like this:

from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, confusion_matrix

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42)
model = DecisionTreeClassifier(max_depth=4, random_state=42)
model.fit(X_train, y_train)
pred = model.predict(X_test)
print(accuracy_score(y_test, pred))
print(confusion_matrix(y_test, pred))

Worked example

A learner builds a model to predict whether a gym member will cancel their membership (yes or no). The cleaned data set has 500 rows, so the 80/20 split gives 400 rows for training and 100 for testing. On the test set the model produces:

• Accuracy = (40 + 45) ÷ 100 = 0.85, or 85%.
• Precision = 40 ÷ (40 + 5) ≈ 0.889: when the model says “will cancel”, it is right about 89% of the time.
• Recall = 40 ÷ (40 + 10) = 0.80: it finds 80% of the members who really cancel.
• F1 score = 2 × 0.889 × 0.80 ÷ (0.889 + 0.80) ≈ 0.84.

Evaluation: if the objective was 80% accuracy, it has been met. But the 10 false negatives are members the gym would fail to contact before they leave. If keeping members matters more than avoiding unnecessary offers, the learner should refine the model to raise recall, accepting a small fall in precision — exactly the kind of reasoned trade-off that supports the aim C Distinction.

Common mistakes that cost marks

• Choosing a project that is not really AI, such as a fixed if-else quiz.
• Copying a tutorial data set and code with no personal objectives or changes.
• Defining the five Vs without applying them to your own data.
• Cleaning data with no before/after evidence.
• Reporting accuracy only, on an imbalanced data set where it hides poor performance.
• Testing on the training data, which gives an unrealistically high score.
• Listing ethical issues without linking them to your project.

How to move from Merit to Distinction

Merit work analyses and refines; Distinction work judges. For the aim A Distinction, weigh benefits against risks across industries and conclude where AI has the greatest positive and negative impact. For the aim B and C Distinctions, return to your objectives and ask: did the solution meet them, how do you know, and what limits the result — data size, bias, model choice? Recommend specific improvements and explain why each would help. Honest evaluation of a modest model scores better than an unsupported claim that a model is “very accurate”.

FAQs

What programming language should I use for Unit 21 AI?

The brief expects Python. Libraries such as pandas for data handling and scikit-learn for models are the simplest starting point; TensorFlow suits image or deep learning projects.

Is my project weak AI or strong AI?

Almost certainly weak (narrow) AI, because it performs one specific task. Strong (general) AI, able to reason across any task like a human, does not yet exist.

Where can I find a data set for my AI project?

Use open government data portals, public data set repositories or data you collect yourself through a survey. Check the licence and remove any personal data you do not need.

What accuracy do I need to pass?

No set accuracy is needed. Assessors look for a working solution, sensible testing and refinement, and honest evaluation against the objectives you set.

Why use an 80/20 train-test split?

Keeping 20% of the data unseen during training lets you check how well the model works on new data and reveals overfitting.

Need one-to-one support from a subject specialist? Our BTEC Level 3 assignment help service covers planning, feedback on drafts and referencing.

If your code will not run or you want feedback on your evaluation, you can get expert help with your BTEC assignment. You may also find our BTEC Level 3 Unit 1 Information Technology Systems answer guide useful.