A novel network-based SIS framework for improved GA performance
Genetic algorithms have long been used to solve complex optimization problems by mimicking natural selection processes. However, they often suffer from premature convergence, reduced diversity, as well as imbalanced exploration and exploitation. To address these challenges, this work introduces SIS-...
Saved in:
| Main Author: | |
|---|---|
| Format: | masterThesis |
| Published: |
2025
|
| Online Access: | http://hdl.handle.net/10725/17282 https://doi.org/10.26756/th.2023.838 http://libraries.lau.edu.lb/research/laur/terms-of-use/thesis.php |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Genetic algorithms have long been used to solve complex optimization problems by mimicking natural selection processes. However, they often suffer from premature convergence, reduced diversity, as well as imbalanced exploration and exploitation. To address these challenges, this work introduces SIS-NGA which integrates the Susceptible-Infected-Susceptible (SIS) epidemic model and Genetic Algorithms within a scale-free network topology, to guide the search for optimal solutions. The SIS model is typically used to capture how infectious diseases spread and evolve within populations. In this model, individuals in a population are represented as interconnected nodes in a network. They can transition between two states, namely susceptible and infected. In analogy, we adapt this formulation to improve the performance of genetic algorithms. We represent the set of possible solutions to a complex optimization problem as interconnected nodes in a scale-free network. We assign fit solutions as infected, with a certain probability. Then, infected nodes can spread their genetic traits to neighboring susceptible nodes through basic genetic algorithm operations within the SIS framework and based on defined probabilities. The proposed approach maintains diversity and delays convergence by promoting promising and optimal solutions. We evaluated SIS-NGA using several benchmark functions, and our results and statistical analyses confirm consistent improvements in solution quality and robustness. |
|---|