TimeGAN-VAE Data Generation for a Malaysian Aviation Supply Chain Stress Index

Abstract

This paper presents a hybrid TimeGAN–VAE–PCA framework for generating synthetic high-frequency time-series data under limited and fragmented reporting conditions in the Malaysian aviation supply chain. The proposed approach combines the temporal modelling capability of TimeGAN with the probabilistic regularisation of a Variational Autoencoder (VAE) and the dimensional interpretability of Principal Component Analysis (PCA) to construct a Supply Chain Stress Index (SCSI). Using aviation logistics data spanning 2017–2024, the framework synthesises consistent multivariate indicators despite irregular reporting frequencies. Experimental results show that the hybrid model achieves lower reconstruction errors (RMSE ≈ 0.012), improves temporal similarity (ACF +42%), and reduces runtime by approximately 50% compared with TimeGAN alone. The resulting SCSI captures real-world disruption dynamics, including the sharp stress surge during the 2020–2021 COVID-19 pandemic and the subsequent recovery period. These findings demonstrate the potential of hybrid generative modelling as a practical tool for monitoring systemic supply-chain stress in data-scarce environments and for supporting data-driven decision-making in aviation logistics.