Ant Colony Optimization and its Potential in Data Mining

ANT COLONY OPTIMIZATION AND ITS POTENTIAL IN DATA MININGBy Ben Degler

OVERVIEWAnt Colony Optimization

How it worksData Mining

ClassificationClustering

ANT COLONY OPTIMIZATION (ACO)Introduced in early 1990’sSocial InsectsSwarm Intelligence

Classifies ants as collaborative agents

Searching for food

WHAT IS AN ANT COLONY?Individual antsSimpleCollective Operation

Food gathering in the optimal way

SEARCHING FOR FOODAnts leave nest Trail formsFollow trails

while they exist

SEARCHING CONTINUEDEfficiencyGuidance

THE ORIGINAL ACO Marco Dorigo Applied to an NP

Complete ProblemApproach

ALGORITHM CHARACTERISTICSAppropriate Problem

RepresentationMove from one city to another

until tour is completedLocal heuristic

Trails buildingTransition Rule

Independent of heuristic value and pheromone level

ALGORITHM CHARACTERISTICSConstraint satisfaction

Forces construction of feasible rules

Fitness FunctionPheromone Update Rule

DATA MINING (DM) Availability Multitude of Possibilities New Associations Two Main Techniques

Classification Clustering

CLASSIFICATION Arrangement The Labeled Model

Labeled sets of data

Specific attributes

MAIN TECHNIQUESDecision TreesAssociation RuleK-Nearest Neighbors AlgorithmArtificial Neural Networks

DECISION TREE

ASSOCIATION RULES“if CONDITION then PREDICTION”

K-NEAREST NEIGHBORS

ARTIFICIAL NEURAL NETWORKS

CLUSTERINGUnsupervised

LearningUnlabeled DataTwo Types

Hierarchical Non-

Hierarchical

HIERARCHICALDendrogramMerging of

Classes

NON-HIERARCHICALFocuses on subclassesUses the k-means algorithm

ACO + DM ACO algorithms in the form of IF-

THENIF(Conditions) THEN(class)Conditions: (term_1) AND

(term_2) AND … AND (term_n)Each term is a triple (attribute, operator, value)EX: <smoke=no>

WEATHER DATASETAre we able to play outside today?

Play{yes, no}Four predicting attributes

Outlook{sunny, overcast, rainy}Temperature{hot, mild, cold}Humidity{high, normal}Windy{true, false}

IF<humidity=normal>THEN<yes>

WEATHER DATASETRule constructionApplying ACO to the problem

Node: <humidity=normal>Edges: Quality of attribute term

Ant constructs a ruleEnds in class term node

<play=yes>Complete path is a constructed rule

WEATHER DATASETPath QualityNode QualityGuidance

ACO + DCAbility to form piles

Cluster dead bodiesSimple and complex movementsProbability of moving itemsPheromone levels

ANT COLONY SIMULATION…

WORKS CITEDIoannis Michelakos, Nikolaos Mallios,

Elpiniki Papageorgiu, Michael Vassilakopoulos, “Ant Colony Optimization and Data Mining: Techniques and Trends”, International Conference on P2P, Parallel Grid and Cloud Computing, IEEE Computer Society, pp. 284-286, 2010.