Two‐step clustering for data reduction combining <scp>DBSCAN</scp> and <i>k</i>‐means clustering

A novel combination of two widely-used clustering algorithms is proposed here for the detection and reduction high data density regions. The density-based spatial applications with noise (DBSCAN) algorithm used regions k-means reduction. iterates while successively decrementing DBSCAN search radius, allowing an adaptive factor based on effective density. demonstrated a physics simulation application, where surrogate model fusion reactor plasma turbulence generated neural networks. training dataset created quasilinear gyrokinetics code turbulent transport calculations in plasmas. set consists inputs derived from repository experimental measurements, meaning there potential risk over-representing specific this input parameter space. By applying to dataset, study demonstrates that can be reduced by ˜20 using algorithm, without noticeable loss accuracy. This also provides way analyzing existing high-dimensional datasets biases consequently reducing them, which lowers cost re-populating space higher quality data.