Impedance cardiography (ICG) is widely recognized as a non-invasive, cost-effective technique for evaluating cardiovascular function by tracking hemodynamic parameters. However, the limited availability of extensive and diverse ICG datasets poses a significant challenge for developing robust machine learning models that detect critical landmarks such as the B, C, and X points. In this paper, we address the scarcity issue by proposing a noise-based augmentation method designed to enrich existing ICG datasets without distorting essential diagnostic features. Specifically, Gaussian noise is systematically injected into the signals, with the noise intensity controlled by a tunable parameter that scales to the signal’s instantaneous magnitude. This approach increases the diversity of training samples, thereby reducing overfitting and enhancing the generalization capabilities of machine learning models. We demonstrate the effectiveness of our method on real ICG data collected in a study examining the effect of caffeine intake on young, healthy non-coffee drinkers. Statistical analyses show that the augmented data retain the core distribution characteristics of the original signals, with minimal shifts in mean and variance. Consequently, our noise-based augmentation strategy enables improved feature extraction and facilitates more reliable performance in time-series classification tasks. The proposed technique can be seamlessly integrated into standard data processing pipelines, making it a viable option for a wide range of biomedical and clinical applications.