Expanding conformal prediction to system identification

System Identification (SysID) is a cornerstone of control engineering, aimed at modeling system dynamics from input–output data. However, SysID approaches often overlook uncertainty quantification (UQ), which is crucial for robust decision-making. To address this gap, we propose Conformal System Identification (C-SysID), a model-agnostic framework for UQ in single-output and multiple-output SysID tasks. C-SysID transforms point predictions of any SysID model into High-Quality (HQ) Prediction Intervals (PIs) for single-output systems or Prediction Regions (PRs) for multiple-output systems, ensuring the true output is captured within the specified confidence level. For single-output systems, we use the Updated Residuals (UR) and Weighted Updated Residuals (WUR) methods to generate valid HQ PIs. In the case of multiple-output systems, we apply the Copula method and introduce the Updated Copula (UC) method, which captures output dependencies to generate HQ PRs. We empirically validate C-SysID on diverse single-output and multiple-output SysID datasets. The statistically significant comparative results show that the WUR method outperforms baseline methods, delivering tight PIs with the desired coverage. For multiple-output systems, our proposed UC method generates PRs that achieve accurate joint coverage while maintaining the compactness of PRs. The results highlight the effectiveness of C-SysID in UQ and enhancing uncertainty-aware predictions for SysID applications.