Publication: Q-learning embedded sine cosine algorithm for general optimization problems
Loading...
Date
2023-04-01
Authors
Qusay Shihab Hamad
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The Sine Cosine Algorithm (SCA) has limitations in solving high-dimensional and complex optimization problems, such as slow convergence rate, tendency to be trapped in local optima, and linear behavior in the exploration and exploitation phases. To overcome these limitations, this study presents a new variant of SCA called Q-learning Embedded Sine Cosine Algorithm (QLESCA), which uses Q-learning to guide search agents to the optimal search area based on previous search history and rewards/penalties received. QLESCA was evaluated and compared to other optimization algorithms, including various SCA variants, on a range of benchmark functions and real-world engineering design problems. The results showed that QLESCA outperformed the other algorithms in terms of fitness and convergence rate. The proposed algorithm was also applied to optimizing a pre-trained Convolutional Neural Network (CNN) for COVID-19 detection from chest X-ray images. The shallow CNN (SCNN) architecture was found to be the most accurate at 96.76%, and QLESCA was used to select a subset of filters for the SCNN to further improve accuracy. This filter selection application of QLESCA resulted in a classification accuracy of 97.22% with 60% of the filters in the first two layers being correctly removed. Additionally, QLESCA was also applied to feature selection for the SCNN model. This application resulted in a classification accuracy of 97.81% with 60% of the total features being eliminated.