Meta-analysis of metaheuristics: Quantifying the effect of adaptiveness in adaptive large neighborhood search

Research on metaheuristics has focused (novel) algorithmic development and competitive testing, both of which have been frequently argued to yield little generalizable knowledge. The main goal this paper is promote meta-analysis — a systematic statistical examination that combines the results several independent studies as more suitable way obtain problem- implementation-independent insights metaheuristics. Meta-analysis widely used in scientific domains, most notably medical sciences (e.g., establish efficacy certain treatment). To best our knowledge, first field illustrate approach, we carry out gain into importance adaptive layer large neighborhood search (ALNS). Although ALNS solve broad range problems, it not yet established whether or adaptiveness actually contributes performance an algorithm. A total 134 were identified through Google Scholar personal e-mail correspondence with researchers domain, 63 fit eligibility criteria. After sending requests for data authors eligible studies, obtained 25 different implementations ALNS, analysed using random effects model. On average, addition algorithm improves objective function value by 0.14% (95% confidence interval 0.06–0.21%). can (and limited number does) added value, also adds complexity therefore only be recommended some specific situations. These findings underline evaluating contribution metaheuristic components.