Hybrid Model Integrating Fuzzy Systems and Convolutional Factorization Machine for Delivery Time Prediction in Intelligent Logistics
Delong Zhu, Zhong Han, Xing Du, Dafa Zuo, Liang Cai, Changchun Xue
Abstract
Logistics distribution plays a crucial role in smart logistics systems, with delivery time prediction being a key issue. Accurately predicting the delivery time of parcels positively impacts logistics companies, couriers, and customers. However, traditional delivery time prediction models struggle to handle potential noise and redundant information in existing features effectively. Moreover, the interaction between various features is often overlooked, and the combination of features is ignored in predicting results. In addition, previous prediction models based on fully connected networks, while capable of learning nonlinear mappings between features, fail to capture local information among features, leading to suboptimal performance on specific datasets. In addition, the intelligent system collects data from multiple sources, introducing ambiguity, which cannot be addressed by fully connected networks in reasoning about the ambiguity among various data. This work proposes a hybrid model using a fuzzy system and convolutional factorization machine (FSCFM) to address the abovementioned challenges. The FSCFM model integrates the local information capture capability of 1-D convolutional neural networks, the automatic feature combination ability of factorization machines, and the reasoning capability of fuzzy systems. This integration allows for a more accurate prediction of delivery times. Multiple experiments were conducted on a real logistics delivery dataset, confirming the practicality of the FSCFM model.