- author = "Marwa Hammami",
- title = "Evolutionary Approaches for Feature Manipulation",
- school = "High Institute of Management, Computer Science Department, University of Tunis",
- year = "2021",
- address = "Tunisia",
- month = "22 " # dec,
- keywords = "genetic algorithms, genetic programming, Feature selection and construction, evolutionary algorithms, multi-objective optimization, bi-level programming, skin cancer diagnosis",
-
URL = "
https://sites.google.com/site/slimbechikh/supervised-phd-theses",
-
URL = "https://drive.google.com/file/d/1UAeG6Vk4laCBI8g4i3SLoXc8FllSTinu/view",
- size = "236 pages",
-
abstract = "Classification tasks often have a large number of
features, but not all of them are useful for
classification. Irrelevant and redundant features may
even decrease the classification accuracy. Feature
selection and feature construction (feature
manipulation) are two important techniques in
preprocessing datasets with irrelevant and not useful
features. While feature selection aims to select
relevant features, feature construction constructs new
features from the original feature set. Both methods
can reduce the dimensionality and increase the
performance of the classification algorithms in terms
of classification accuracy and computation
time.
Existing methods for feature manipulation usually face the problem of high computational time and stagnation in local optima. Evolutionary computation (EC) techniques are well-known global search algorithms. Genetic algorithm (GA) and genetic programming (GP) have shown promise in feature selection and feature construction, respectively. GA and GP have been successfully applied to many areas, but their potential for feature selection and construction has not been fully investigated.
The overall goal of this thesis is to investigate new evolutionary multi-objective approaches to using GA for feature selection and GP for feature construction in classification problems. The proposed methods can be classified in four main classes: (1) new evolutionary multi-objective feature selection method, (2) evolutionary feature construction methods, (3) feature weighting methods for feature selection and construction, and (4) feature manipulation method for skin cancer image classification. First, we propose new evolutionary feature selection approach based on local search using three filter-wrapper objectives. Second, we propose a new evolutionary approaches for feature construction to reduce the dimensionality and improve the classification performance. Then, we propose feature weighting methods for feature manipulation based on a bi-level model. Finally, we propose new methods to skin image classification by using bi-level optimisation and class dependent feature construction, which can help dermatologist to diagnose a type of cancer. All the proposed contributions have been assessed through experimental studies including comparative experiments against the most prominent recent works by using benchmark datasets of varying difficulty. These datasets are commonly used in the literature to evaluate the performance of feature selection/construction methods.",
-
notes = "Research laboratory: strategies for modeling and
artificial intelligence (SMART) University of Tunis,
ISG
Supervisor: Slim Bechikh",
