Exploring Experts Decisions in Concrete Delivery Dispatching Systems Using Bayesian Network Learning Techniques

Abstract

Optimally solving large scale Ready Mixed Concrete Dispatching Problems (RMCDPs) in polynomial time is a crucial issue and, in the absence of automated solutions, experts are hired to handle resource allocation tasks in concrete dispatching centres. Therefore, in the Ready Mixed Concrete (RMC) industry, the performance of experts is accepted as the only practical solution, although there is no benchmark for assessing the quality of their decisions. This paper aims to discover the experts' decisions in the RMC context by using Bayesian Network. Finding the optimum graph in Bayesian Network is NP-hard, therefore, this research uses a wide range of heuristic search algorithms (Hill Climbing, K2, Look Ahead Hill Climbing, Repeated Hill Climbing, Tabu Search, Simulated Annealing and Genetic Algorithm). A large scale dataset gathered from an active RMC was used for evaluating the proposed idea. Results show that Simulated Annealing search algorithm outperformed other search algorithms, although there is not a significant difference between them. However, interpreting the network obtained by Simulated Annealing involves much more effort than other networks with similar accuracy, such as K2.