First NameArindam
Last NameChaudhuri
Emailarindamphdthesis@gmail.com
Supervisor NameProf Kajal De
UniversityNetaji Subhas Open University
CountryIndia
KeywordsSoft Computing, Optimization Problems, Uncertainty, Vagueness, Impreciseness, Fuzziness, Optimal Solutions
Publication Date24 September, 2017
DegreePhD
DomainComputer Science / IT

Studies in Application of Soft Computing in Optimization Problems

Abstract: The ever increasing demand to lower production costs to withstand cutthroat competition has prompted engineers and technologists to look for rigorous methods of decision making to design and produce economic and efficient products. As search for the best has always fascinated mankind, operations or strategies have been attempted and devised for searching the optimum solution of variety of problems in almost all branches of activities that can be perceived by logic or intitution or both. With this motivation in this thesis novel Soft Computing tools are proposed to develop optimum solutions for different aspects of optimization problem. The problems considered in this research work are traveling salesman problem, transportation problem, decision making problem, rectangular game problem, financial investments problem, stock price prediction problem, time series forecasting problem, bankruptcy prediction problem, resource allocation problem, assignment problem, sequencing and job scheduling problem. The various methodologies are developed using Soft Computing approaches by integrating fuzzy logic, artificial neural networks, rough sets, genetic algorithms and ant colony optimization. The emphasis of proposed methodologies is given on handling data sets which are large both in size and dimension and involves classes that are overlapping, intractable and have non-linear boundaries. Several strategies based on data reduction, dimensionality reduction, active learning efficient search heuristics are employed for dealing with the issue of scaling in learning problem. The problems of handling linguistic input and ambiguous output decision, learning of overlapping and intractable class structures, selection of optimal parameters and discovering human comprehensible knowledge in form of linguistic rules are addressed in the Soft Computing framework. The different features of methodologies along with comparisons with those of related ones are demonstrated extensively on different real life data sets. The experimental data have number of dimensions and are considered from varied domains such as traveling salesman problem, banking systems, financial institutions, corporate organizations, stock exchanges and currency exchange rates, academic institutions, scheduling and sequencing problems. The superiority of models over the benchmark are found to be effective and significant.

Download Thesis