Sequential Covering and Ant Miner

Question 1

What 2 ways can ML be used to optimise algorithms?

Answer 1

Model parameters -> ML can be used to model parameters.

Algorithm -> Can be used to optimise the algorithm itself.

Question 2

What is rule-based classification?

Answer 2

• A technique for creating a classification model by deriving classification rules from data.

Question 3

Explain how Sequential Covering works…

Answer 3

• An algorithm for creating a classification model by deriving classification rules from discrete data.
• Processes data and iteratively adds rules to a set of learned rules. These learned rules can then be used for classification on unseen data instances.

Question 4

What is the benefit of using Ant Miner over sequential covering?

Answer 4

• EAs are better at dealing with large data.
• EAs can operate in parallel over a large search space E.g ACO.

Question 5

Explain the process of Ant Miner

Answer 5

1. Create the Ant Miner graph. Create node for every attribute and perform discretization as needed.
2. Connect all attributes except for ones of the same type.
3. Send ants along graph
4. Assess fitness and adjust pheromones
5. Repeat until termination criteria
6. Select a set of classification rules (paths) with greatest pheromone values.

Question 6

Explain what the discretization step is and why it’s needed…

Answer 6

The process of transforming continuous data into discrete data.

This enables Ant Miner to operate on continuous data.