AI Project Cycle

The AI project cycle has six stages: problem scoping, data acquisition, data exploration, modelling, evaluation and deployment.

Key ideas

Problem scoping uses the 4Ws canvas to say clearly who has the problem, what it is, where it happens and why solving it helps.
Data acquisition collects reliable data on the features that affect the problem, such as rainfall for crop yield.
Data exploration uses graphs and summaries to find patterns, gaps and odd values before modelling.
A rule-based model follows rules written by people; a learning-based model finds its own rules from data.
Labelled data is split into a training set for learning and a test set kept back for evaluation.
Deployment puts the model into real use, in an app or a system, where it must keep being checked.

Remember

The six stages	problem scoping → data acquisition → data exploration → modelling → evaluation → deployment
4Ws	Who is affected? What is the problem? Where does it arise? Why will solving it help?
Train : test split 80 : 20 of 500	400 for training, 100 for testing
Readings in a day	1,440 minutes ÷ minutes between readings
Accuracy on the test set	correct ÷ test examples × 100%

Watch out

Jumping to modelling before the problem is scoped and the data explored.
Testing on the training data, which makes a model look better than it is.
Calling a fixed-rule program learning-based.

Worked example

Put the stages of the AI project cycle in order. Write the letters in order.
(a) Problem scoping
(b) Deployment
(c) Data acquisition
(d) Data exploration
(e) Evaluation
(f) Modelling

Working: Problem scoping → Data acquisition → Data exploration → Modelling → Evaluation → Deployment

Answer: a, c, d, f, e, b