A fitness landscape analysis of the travelling thief problem
Mohamed El Yafrani, Marcella Scoczynski Ribeiro Martins, Mehdi El Krari, Markus Wagner, Myriam Delgado, Belaïd Ahiod, Ricardo Lüders · Proceedings of the Genetic and Evolutionary Computation Conference · 2018
Local Optima Networks are models proposed to understand the structure and properties of combinatorial landscapes. The fitness landscape is explored as a graph whose nodes represent the local optima (or basins of attraction) and edges represent the connectivity between them. In this paper, we use this representation to study a combinatorial optimisation problem, with two interdepend components, named the Travelling Thief Problem (TTP). The objective is to understand the search space structure of the TTP using basic local search heuristics and to distinguish the most impactful problem features. We create a large set of enumerable TTP instances and generate a Local Optima Network for each instance using two hill climbing variants. Two problem features are investigated, namely the knapsack capacity and profit-weight correlation. Our insights can be useful not only to design landscape-aware local search heuristics, but also to better understand what makes the TTP challenging for specific heuristics.