Improving artificial algae algorithm performance by predicting candidate solution quality

• Published in: Expert systems with applications Vol. 150; p. 113298
• Main Authors: Yibre, Abdulkerim Mohammed, Koçer, Barış
• Format: Journal Article
• Language: English
• Published: New York Elsevier Ltd 15.07.2020; Elsevier BV
• Subjects: Algorithms; Artificial algae algorithm; Candidate solution prediction; Fitness; Gaussian distribution; Heuristic methods; Naïve Bayes; Optimization; Optimization algorithms; Performance prediction; Probabilistic models; Swarm intelligence; Artificial algae algorithm; Naïve Bayes; Swarm intelligence; Candidate solution prediction
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2020.113298

•The proposed method can find better results with same maximum fitness evaluations.•Candidate solution prediction boosted the performance of artificial algae algorithm.•The proposed method has shown better convergence rate than 7competitor algorithms.•The new method contributed for better balance of exploration and exploitation.•Probabilistic model helped to achieve better result with fewer function evaluations. The success of optimization algorithms is most of the time directly proportional to the number of fitness evaluations. However, not all fitness evaluations lead to successful fitness updates. Besides, the maximum number of fitness evaluations is limited and also balance of exploration and exploitation is still challenging. Best possible solution should be found in a reasonable time. Surely it can be said more fitness evaluation takes more time. Since methods are tested under fixed numbers of maximum fitness evaluation and the duration of each fitness evaluation of a problem may vary depending on the characteristic of the problem, finding best result with fewer fitness evaluations is challenging in optimization algorithms. For that reason in this study, we proposed a new method that predicts the quality of a candidate solution before evaluation of its fitness employing Gaussian-based Naïve Bayes probabilistic model. If the candidate solution is predicted to generate good result then that solution is evaluated by the objective function. Otherwise new candidate solution is created as usual. The primary purpose of the proposed method is improving the performance of AAA and at the same time preventing unnecessary fitness evaluation. The proposed method is evaluated using standard benchmark functions and CEC’05 test suite. The obtained results suggests that the new method outperformed the basic AAA and other state-of-the-art meta-heuristic algorithms with fewer fitness evaluations. Thus, the new method can be extended to cost sensitive industrial problems.