TutorSyntax — Learn. Practise. Build.
TutorSyntax · Learn. Practise. Build.

AI & Data Science Foundations · Answer Guidance

Examples of evidence-based responses for practical projects in AI & Data Science Foundations. These are illustrative explanations, not an official mark scheme or a substitute for your own investigation.

Book chapters and files →All book solutions →
Use these as models, not scripts to copy. The book contains suggested answers and teaching guidance. For open-ended investigations, the strongest response depends on accurate data, clearly described methods, honest limitations and ethical reflection. Reveal the example response only after analysing the evidence.
Chapter 17 · Analyse a Real Dataset

How can I check an Iris dataset conclusion?

The real Fisher Iris companion file has 150 records, four measurements (centimetres) and three species. It is evidence for comparisons between these flowers, not for claiming that one measurement causes a species.

1 · Reveal an investigation checklist
  1. Check all 150 records load and the species labels are consistent.
  2. Calculate and label mean petal length in centimetres for each species.
  3. Compare overlap and variation, not just the averages.
  4. State a limitation: this historical sample is not every iris flower.
2 · Show an illustrative model response
“The sample shows differences in petal lengths between the species, but some measurements overlap. The observed pattern helps comparison; it does not establish a causal explanation or guarantee classification of every new flower.”

Other evidence-based conclusions may also be valid.

Check and evaluate: Repeat one calculation by hand and check units; use two visualisations if possible.

Download the matching chapter files ↓

Chapter 18 · Build a Simple Prediction System

What counts as an evidence-based prediction answer?

The fictional bike-hire file contains 24 daily observations. The book uses days 1–16 for training and days 17–24 for evaluation.

1 · Reveal an investigation checklist
  1. Keep all eight test rows unseen while constructing the model.
  2. Predict bike hires and compare predicted and actual values.
  3. Calculate mean absolute error (MAE) over the same eight test rows.
  4. Compare against a simple baseline and discuss rain, season and other omitted factors.
2 · Show an illustrative model response
“The model can estimate demand from temperature on these fictional examples, but it is not certain. A test MAE alone does not prove it will perform well in another place or season; we should compare it with a baseline and check relevant omitted factors.”

Other evidence-based conclusions may also be valid.

Check and evaluate: Do not calculate training and testing accuracy using the same records. Include the exact test count and units (hires).

Download the matching chapter files ↓

Chapter 20 · Final Portfolio Project and Progression

How do I write a reproducible sustainability conclusion?

Project A uses fictional café counts. The five printed sample days have 244 disposable containers out of 500 (48.8%); the complete ten-day CSV has 458 out of 1,000 (45.8%).

1 · Reveal an investigation checklist
  1. State whether you analysed the printed five-day example or the entire ten-day CSV.
  2. Give numerator, denominator and percentage, not just a percentage.
  3. Create a clearly labelled chart and document your checking method.
  4. Suggest a cautious action and two limitations, rather than claiming a trend or an intervention outcome.
2 · Show an illustrative model response
“In the ten-day fictional dataset, 458 of 1,000 recorded containers were disposable (45.8%). This measures use over those ten days only; it does not prove a trend or which change would reduce waste. Further evidence would be needed.”

Other evidence-based conclusions may also be valid.

Check and evaluate: Five-day and ten-day results are different periods; both figures are correct for their respective datasets.

Download the matching chapter files ↓

Portfolio quality checklist

For any AI or data investigation, aim to show a specific question, source and permissions, defined observation period, data-quality checks, repeatable calculations, labelled outputs, limitations, and a reflection on improvements.

In particular, never treat a model prediction as a guarantee or upload identifiable student information to unapproved services.

Try the interactive Data Science Studio →Explore independent research planning →