rules reduction using evolutionary meta-heuristics
TRANSCRIPT
Rules Reduction using
Evolutionary Meta-Heuristics
*Hanaa Ismail Elshazly and Aboul Ella Hassanien
http://www.egyptscience.net
*Faculty of Computers and Information, Cairo University, and SRGE member
Bio-inspiring and evolutionary computation: Trends, applications and open issues workshop, 7 Nov. 2015 Faculty of Computers and Information, Cairo University
Overview
IntroductionProblem DefinitionMotivation
Background Proposed ApproachResults and DiscussionConclusion and Future Works
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
2
Introduction
3
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Problem Definition
Decision rules suffer of some shortcomings like Difficulty of knowledge acquisition process. Big size of rules that need maintenance. More cost development and time from experts and
knowledge engineers.
The need for an automatic system for good decision rules selected will offer great help and will reduce cost and effort for knowledge base construction and rules maintenance.
4
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Motivation In Biomedical Field
Medical diagnosis depends on the experience of the physician, Decision rules can transfer this experience.
Decision rules present an easy and strong method of inference consistent with expert knowledge and ability of expression and explanation.
Decision rules can be easily adapted due to its declarative representation.
5
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Background
6
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Background
Meta-Heuristic Optimization Techniques have been applied in various fields of study.
It can divided into Single-solution-based and Population-based.
All meta-heuristic techniques divided the search space into two phases : exploration and exploitation.
7
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Background Meta-Heuristic Optimization techniques can be
divided into :
Evolutionary Concepts : Inspired from evolutionary processes and
natural operations like crossover, mutation and selection, GA most
popular.
Physical Phynomena : Random set of search agents
communicate and move throughout search space according to physical
rules like gravitational force and ray casting.
Animal Behaviours : Imitate social behavior of swarms, herds,
schools of creatures in nature, using the simulated collective and social
intelligence of creatures.
8
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Background
GA simulates Darwnian evolution concepts.
Optimization is done by evolving an initial random
solution.
Each new population is created by the combination
and mutation of the individuals in the previous
generation.
Best individuals are chosen and participate in
generating new population.
9
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Proposed
Approach
10
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Proposed Approach11
+
Preprocess
Normalized Data
Breast Cancer
Data
200 Reducts &
100 000 Rules
RS
Evaluation
Process
Selected Rules
Evolution
algorithms
Reduced Rules
Selected Reduct
Generate
Rules
Rules Rendring
Classified
Instances
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and
Discussion
12
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion13
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion14
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion15
Visualization of 1000 rules
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion16
Visualization of 500 rules
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion17
Visualization of 98 rules
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion18
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion19
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Results and Discussion20
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Conclusion and Future Works
21
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Conclusion
In this study, a hybrid method for breast cancerdiagnosis based on RS for feature selection and GAmetaheuristic is presented.
The hybrid proposed method exhibits consistent andbetter performance than the RS classifier only.
The hybrid proposed method provides the expert by apromising tool to extract knowledge visually andinstantly.
The hybrid proposed method represents anindependent phase for classification and reductionwhich can recover previous steps defects.
22
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)
Future Works
More biomedical data sets with various
dimensionality will be tested.
Search for unique criterions within each
medical domain to improve selection rules
process visualization.
Test other optimization techniques for rules
reduction.
23
SRGE Workshop, Bio-Inspiring and Evolutionary Computation Trends: Applications and Open Issues (7-November-2015)