Scholarlink

<p>The Capacitated Vehicle Routing challenge is a typical hard combinatorial optimization problem for which numerous algorithmic techniques have been developed. Empirical studies have consistently shown that there is no universally accepted optimal strategy that works equally well on all instances of a problem. It indicates that performance of an algorithm varies across different scenarios. Majorly, performance of algorithms depends on structural and other properties of problem instance. The challenge of identifying the most effective solution method from a predefined set of options is addressed through algorithm selection problem. To this end, a meta-learning-based categorization system is proposed. This study employs a heuristic algorithm to a collection of instances, categorising them as either ‘solved’ or ‘not solved’, depending on whether the algorithm successfully solves the instance inside specific configuration. The label created along with the characteristics of instance provides a meta example in the meta-data. A set of binary classifiers are learned to predict the suitability of algorithm for different instances. Experimentation is done on 5000 CVRP instances. Eight different classifiers demonstrate accuracy and other performance metrics up to 90%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Meta-learning-Based Model to Select Algorithm for Capacitated Vehicle Routing Problem

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Neha Sehta,
  • Urjita Thakar

摘要

The Capacitated Vehicle Routing challenge is a typical hard combinatorial optimization problem for which numerous algorithmic techniques have been developed. Empirical studies have consistently shown that there is no universally accepted optimal strategy that works equally well on all instances of a problem. It indicates that performance of an algorithm varies across different scenarios. Majorly, performance of algorithms depends on structural and other properties of problem instance. The challenge of identifying the most effective solution method from a predefined set of options is addressed through algorithm selection problem. To this end, a meta-learning-based categorization system is proposed. This study employs a heuristic algorithm to a collection of instances, categorising them as either ‘solved’ or ‘not solved’, depending on whether the algorithm successfully solves the instance inside specific configuration. The label created along with the characteristics of instance provides a meta example in the meta-data. A set of binary classifiers are learned to predict the suitability of algorithm for different instances. Experimentation is done on 5000 CVRP instances. Eight different classifiers demonstrate accuracy and other performance metrics up to 90%.