Abstract <p>Imaging atmospheric Cherenkov telescopes (IACT) are commonly used in high-energy ground-based gamma-ray astronomy to detect the angular distribution (images) of Cherenkov light from extensive air showers (EAS). Using these images, it is necessary to determine the arrival direction, energy, and type of the primary particle. Particle identification is crucial for extracting the gamma signal from cosmic-ray background. Machine learning (ML) methods, particularly artificial neural networks (ANN), enable highly accurate estimation of primary particle parameters. However, training ANN requires large labeled datasets generated through EAS and telescope system simulations. EAS simulation is computationally intensive; a single shower modeling may take several hours. Thus, this limitation restricts the creation of large datasets, making data augmentation essential. The issue becomes particularly acute in stereo mode, as the number of coincident events decreases dramatically with an increase in the number of telescopes registering the same event. Simple image rotation for monomode data augmentation has proven effective. This study explores rotating telescope positions around the EAS axis to augment stereo-mode data for ANN training. The research is carried out for the TAIGA experiment, which includes IACTs.</p>