Transcript Medical Data Mining Using Fuzzy Evolutionary Computing

Medical Data Mining Using
Fuzzy Evolutionary Computing
Abdul Razak Hussain
1 October 2002
Contents
•
•
•
•
Introduction
Literature Review
Methodology
Summary
Introduction
•
•
•
•
Overview
Background of problem
Statement of problem
Purpose of study
Overview
• Medical/Biological databases
– broad range: dermatological, gynaecological,
neurological, urological, obstetrical, immunological,
etc.
– vast data: 200,000 records with 60 fields, may spans
decades (DW)
– decision-making: operationally or strategically (DM)
Background of problem
• Heart disease
• Evolutionary computing & medical data
Heart disease
• 25 % of death rates in the world (developed countries).
• World Heart Federation: “one out of three deaths across
the world are now due to heart disease and stroke”.
• WHO: by 2025, heart disease is the leading killer (noncommunicable disease) – more common in developing
countries.
• Kementerian Kesihatan Malaysia: heart diseases &
diseases of pulmonary circulation are leading causes of
deaths in govt. hospitals (1996-1999).
Heart disease..2
• Cardiovascular disease (CVD) is the main cause of
death in the UK accounting for over 235,000 deaths a
year: around four out of ten of all deaths. The main
forms of CVD are coronary heart disease (CHD) and
stroke. About half of all deaths from CVD are from CHD
and about a quarter are from stroke.
(source: http://www.heartstats.org/)
Heart disease..3
• In 1999 CVD contributed to one-third of global deaths.
Low- and middle-income countries contributed to 78
percent of CVD deaths.
• By 2010 CVD is estimated to be the leading cause of
death in developing countries. Heart disease has no
geographic, gender or socioeconomic boundaries.
(Source: American Heart Association)
Evolutionary computing & medical data
• optimisation, automatic programming, circuit
design, machine learning, economics, ecology and
population genetics
• Solving medical problems: expert systems, ANN,
evolutionary computing (EC)
• diabetes, breast cancer, chest pains, preterm
births, cervical cancer, ovarian cancer
Statement of problem
• As a crucial and important procedure in
medicine, diagnosis of diseases need to be
performed with acceptable accuracy.
• Data mining technique incorporating fuzziness in
evolutionary computing methods, particularly in
genetic algorithm, still lacks.
• To develop a hybrid evolutionary algorithm to
diagnose heart disease.
Purpose of study
• To develop a hybrid fuzzy genetic algorithm
(FGA) capable of detecting heart disease based
on available data sets.
• To compare the performance of our algorithm
with other selected hybrid systems.
• To test the proposed algorithm in the Malaysian
context.
Literature Review
• KDD - selection, cleaning, enrichment, coding, DM,
reporting
• Data mining (DM) overview
–
–
–
–
Classification
Clustering
Association
Sequences
Literature Review..2
• Evolutionary computing (EC):
–
–
–
–
genetic algorithm (GA) – (Holland 1975)
genetic programming (GP) – (Koza 1992)
evolution strategies (ES) – (Rechenberg 1973)
evolutionary programming (EP) – (Fogel, Owens,
Walsh 1966)
• Soft computing (SC) = EC + ANN + FL
Methodology
• Research - experimental studies
• Research design:
– development of governing equations to model various heart
diseases.
– selection and setting-up of relevant parameters required for
the development of a hybrid fuzzy genetic algorithm (FGA).
– planning of the framework in the development of the hybrid
FGA.
– coding of the hybrid FGA using C++/Java
– testing of the algorithm using historical data.
– comparison of the performance of the algorithm
– testing of the algorithm using current data.
Methodology..2
• Planning and execution
– May 2002 until June 2005
• Data sources and instrumentation
• Assumptions and limitations
• Expected results
Summary
• A novel hybrid fuzzy evolutionary computing
algorithm that is capable of detecting heart disease,
in the hope that it will achieve a better rating than
other hybrid algorithms.
“Evolution is the natural way to program” -Thomas Ray