<p>Revealing the connectivity patterns of complex oscillatory networks is essential for understanding their dynamical behavior. However, direct measurement of large-scale network connections is often infeasible, making inference methods based on observed time-series data indispensable. Accurate network reconstruction is typically hindered by synchronization, noise, and dense connectivity. In this study, we introduce Local-Deviation-based Iterative Sparse Regression, a method that integrates iterative local linearization with sensitivity-guided refinement of local neighborhoods, thereby improving the stability and accuracy of network inference, particularly in systems exhibiting transient or oscillatory behavior. We evaluate its effectiveness on Hodgkin–Huxley and Izhikevich neuronal networks and apply it to reconstruct coupled Lorenz oscillator networks, thereby illustrating its versatility across different classes of dynamical systems.</p>