AI Project Lifecycle

An AI project goes through six stages, from defining the problem to monitoring the model in use. Good data and fair testing decide how well it works.

Key ideas

First decide whether the task really needs AI, which learns and adapts, or only fixed automation.
Data comes from sensors, surveys, websites and historical records; it must be cleaned before training.
Cleaning removes repeated rows and deals with missing values; labelling marks each example with its answer.
A model learns patterns from the training data, then is tested on new data it has never seen.
After deployment the model is monitored and updated when its accuracy drops or conditions change.
A learned rule like “on time only if every condition holds” is applied to each new row in turn.

Remember

The six stages, in order	defining the problem → data collection and preparation → model development and training → model evaluation and refinement → model deployment → monitoring and maintenance
Accuracy	correct ÷ number tested × 100%
Fewest right for p% accuracy	p% of the test set: 80% of 25 = 20
Readings in a day	1,440 minutes ÷ minutes between readings

Watch out

Testing a model on the same data it was trained on.
Mixing up evaluation (checking on new data) and monitoring (checking after it is in use).
Counting the first appearance of a name as a repeat when cleaning data.

Worked example

Put the stages of the AI project lifecycle in order. Write the letters in order.
(a) Data collection and preparation
(b) Defining the problem
(c) Monitoring and maintenance
(d) Model evaluation and refinement
(e) Model deployment
(f) Model development and training

Working: Defining the problem → Data collection and preparation → Model development and training → Model evaluation and refinement → Model deployment → Monitoring and maintenance

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