A Proposed Feature Selection Particle Swarm Optimization Adaptation for Intelligent Logistics—A Supply Chain Backlog Elimination Framework

Abstract

The diversity of data sources resulted in seeking effective manipulation and dissemination. The challenge that arises from the increasing dimensionality has a negative effect on the computation performance, efficiency, and stability of computing. One of the most successful optimization algorithms is Particle Swarm Optimization (PSO) which has proved its effectiveness in exploring the highest influencing features in the search space based on its fast convergence and the ability to utilize a small set of parameters in the search task. This research proposes an effective enhancement of PSO that tackles the challenge of randomness search which directly enhances PSO performance. On the other hand, this research proposes a generic intelligent framework for early prediction of orders delay and eliminate orders backlogs which could be considered as an efficient potential solution for raising the supply chain performance. The proposed adapted algorithm has been applied to a supply chain dataset which minimized the features set from twenty-one features to ten significant features. To confirm the proposed algorithm results, the updated data has been examined by eight of the well-known classification algorithms which reached a minimum accuracy percentage equal to 94.3% for random forest and a maximum of 99.0 for Naïve Bayes. Moreover, the proposed algorithm adaptation has been compared with other proposed adaptations of PSO from the literature over different datasets. The proposed PSO adaptation reached a higher accuracy compared with the literature ranging from 97.8 to 99.36 which also proved the advancement of the current research.