Lean Yu

The rapid growth of on-demand meal delivery has increased the importance of improving delivery efficiency while effectively managing the rider workforce. While prior studies have examined various operational factors affecting delivery outcomes, limited attention has been paid to how riders learn from accumulated delivery experience. Drawing on the learning-by-doing perspective, we examine the effect of rider experience on delivery time and explore the moderating effects of peak periods, traffic congestion, and rider workload (i.e., the number of outstanding orders). Econometric analyses of order-level operational data from an on-demand meal delivery platform reveal a negative, nonlinear relationship between rider experience and delivery time, with significant diminishing marginal performance gains. More importantly, we find that peak periods and higher rider workload strengthen and attenuate the learning effect, respectively, while the strengthening effect of traffic congestion is only partially supported across model specifications. These findings suggest that the performance benefits of delivery experience depend not only on accumulated experience itself but also on operating conditions, offering valuable implications for understanding rider learning and designing experience-aware dispatching and workforce management policies.

Financial fraud detection is crucial in the banking and financial sectors, but it faces considerable challenges because of class imbalance in transaction data and the growing threat of adversarial attacks. These issues frequently undermine the effectiveness of deep neural networks despite their demonstrated potential in this domain. To address these challenges, this paper proposes a novel approach, adversarial training–based deep imbalanced learning (ATDIL), which integrates imbalanced learning and adversarial defense into a unified approach. ATDIL leverages an adversarial autoencoder to efficiently synthesize high-quality minority-class samples that are informative and adversarial, while maintaining low computational complexity. The effectiveness of ATDIL is rigorously validated through both theoretical and experimental analyses. Theoretically, the optimal solution form for the inner optimization problem in ATDIL is derived, and its convergence under mild assumptions is established. Extensive evaluations on seven real-world financial data sets demonstrate that ATDIL outperforms state-of-the-art imbalanced learning methods across multiple metrics while exhibiting superior resilience under adversarial conditions. This combination of theoretical guarantees and empirical evidence highlights ATDIL’s ability to effectively address class imbalance and model security, offering a robust and practical framework for enhancing financial fraud detection systems. History: Accepted by Ram Ramesh, Area Editor for Data Science and Machine Learning. Funding: This work was supported by the National Natural Science Foundation of China [Grant 72401208], the Key Program of the National Natural Science Foundation of China [Grant 72331007], the Natural Science Foundation of Sichuan Province [Grant 2025NSFSC1981], the Postdoctoral Fellowship Program of the China Postdoctoral Science Foundation [Grant GZB20240504], the International Visiting Program for Excellent Young Scholars of Sichuan University (SCU), and the Humanities and Social Science Youth Foundation of the Ministry of Education of China [Grant 23YJCZH088]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2025.1251 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2025.1251 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

In the field of credit risk classification, the prediction model needs to meet a large number of labelled samples to make the model fully trained. However, in some practical scenarios, it is difficult for financial institutions to obtain a large number of available labelled samples because of the cost or privacy. Shortage of samples will cause the prediction model ineffective or even invalid. To address low sample size problem, a novel similarity‐based graph neural network with meta‐learning called ML‐SGNN is proposed for credit classification with low sample size. In the proposed ML‐SGNN methodology, attribute co‐training based on meta‐learning strategy is first introduced to increase the generalisation capability of the proposed model under label scarcity condition. Then, a node update strategy based on similarity is proposed to aggregate features of samples with same label so as to provide a contrastive signal that increases the discriminative information extracted from low sample size. Finally, a novel similarity‐based edge update strategy utilising the difference of node features and convolutional neural network is proposed to better describe the similarity relationships among unordered attribute structures to improve the generalizability of ML‐SGNN. The experiment results demonstrate that the proposed ML‐SGNN method consistently achieves the highest average ranking across different sample sizes and evaluation metrics among the compared methods. Generalisation capability of ML‐SGNN indicating that ML‐SGNN can be considered as a promising solution for credit classification with low sample size.

The software and data in this repository are a snapshot of the software and data that were used in the research reported on in the paper Adversarial Training-Based Deep Imbalanced Learning by Yuhang Tian, Jin Xiao, Lean Yu, Yanlin Jia, Jing Huang, Xiaoyi Jiang, and Shouyang Wang.