A Comprehensive Review on NSGA-II for Multi-Objective Combinatorial Optimization Problems

Analysis

Why This Paper Matters

This review paper is significant because NSGA-II is one of the most widely used multi-objective evolutionary algorithms, yet its application to combinatorial optimization problems is often ad hoc. By systematically categorizing implementations into conventional, modified, and hybrid variants, the authors provide a clear framework for understanding how NSGA-II has been adapted for problems like scheduling, vehicle routing, and the knapsack problem. This taxonomy helps practitioners quickly identify which type of variant might be suitable for their specific problem and what modifications have been effective.

The inclusion of a bibliometric analysis adds a meta-perspective, revealing research trends, influential papers, and collaboration patterns. This is valuable for researchers looking to understand the evolution of the field and identify gaps or emerging directions. The paper also consolidates information on performance assessment techniques, which is often scattered across individual studies.

Technical Contributions

• Three-category taxonomy: Conventional NSGA-II (no operator changes), Modified NSGA-II (changes to crossover, mutation, or selection), and Hybrid NSGA-II (combined with local search, other metaheuristics, or exact methods).
• Problem-specific analysis: For each of the six problem types (assignment, allocation, TSP, VRP, scheduling, knapsack), the review summarizes how NSGA-II has been applied and which modifications were common.
• Performance assessment review: Compilation of test instances, performance metrics (e.g., hypervolume, generational distance, inverted generational distance), statistical tests, and benchmarking practices used in the literature.
• Bibliometric analysis: Co-authorship networks, citation analysis, and keyword co-occurrence to map the research landscape.

Results

The paper does not present new experimental results but synthesizes findings from 930 cited works. Key observations include: (1) Hybrid NSGA-II variants are the most common, often combining NSGA-II with local search or other heuristics to improve convergence. (2) Modified NSGA-II frequently involves adaptive crossover/mutation rates or problem-specific encoding schemes. (3) Performance metrics like hypervolume and inverted generational distance are standard, but many studies lack statistical significance testing. The bibliometric analysis shows increasing publication volume and a shift toward hybrid approaches in recent years.

Significance

This review serves as a comprehensive reference for AI practitioners working on multi-objective combinatorial optimization. By clarifying the landscape of NSGA-II variants, it reduces the time needed to select an appropriate algorithm and highlights best practices for performance evaluation. The taxonomy also encourages more systematic comparisons and could inspire new hybrid or modified variants. For the broader AI field, the paper underscores the importance of adapting general-purpose algorithms to specific problem structures, a theme relevant beyond evolutionary computation.