An Improved Yin-Yang-Pair Optimization Algorithm Based on Elite Strategy and Adaptive Mutation Method for Big Data Analytics

Abstract

In order to analyze and explore the potential value of big data more effectively, intelligent optimization algorithms are applied to this field increasingly. However, data often has the characteristics of high dimension for big data analytics. As the dimension of data increases, the performance of the optimization algorithm degrades dramatically. The Yin-Yang-Pair Optimization (YYPO) is a lightweight single-objective optimization algorithm, which has stronger competitive performance compared with other algorithms and has significantly lower computational time complexity. Nevertheless, it also suffers from the drawbacks of easily falling into local optimum and elitism deficiency, resulting in unsatisfactory performance on high-dimensional problems. To further improve the performance of YYPO in solving high-dimensional problems for big data analytics, this paper proposes an improved Yin-Yang-Pair Optimization based on elite strategy and adaptive mutation method, namely CM-YYPO. First, the crossover operator using an elite strategy is introduced to record the individual optimal generated by point P₁ as elite. After the splitting stage, the elite to cross-disturb point P₁ is utilized to improve the global search performance of YYPO. Subsequently, the mutation operator with an improved adaptive mutation method is used to mutate point P₁ to improve the local search performance of YYPO. The proposed CM-YYPO is evaluated by 28 test functions used in the Single-Objective Real Parameter Algorithm competition of the Congress on Evolutionary Computation 2013. The performance of CM-YYPO is compared with YYPO, YYPO-SA1, YYPO-SA2, A-YYPO, and four other single-objective optimization algorithms with superior performance, which are Salp Swarm Algorithm, Sine Cosine Algorithm, Grey Wolf Optimizer and Whale Optimization Algorithm. The experimental results show that, the proposed CM-YYPO can achieve more stable optimization capability and higher computational accuracy on high-dimensional problems, which prospects a promising idea to solve high-dimensional problems in the field of big data analytics.