Detection and Analysis of Quasar Spectroscopic Anomalies

We conduct anomaly detection on the spectral data of 26,818 quasars within the redshift range $1.97 \leq z leq 2.16$ from the SDSS DR16Q Quasar catalog. This study encompasses two datasets: one including Broad Absorption Line (BAL) quasars and the other excluding them, facilitating a comparative analysis. Our methodology highlights five categories of peculiar quasars and visually identifies eight truly bizarre QSOs of unknown physical understanding. Notably, our algorithm serendipitously detects and segregates FeLoBAL and LoBAL quasars (subgroups of BAL quasars that are notoriously challenging to identify) into two distinct groups. Additionally, we curate a list of spectra corrupted by noise or artifacts that require SDSS revision and present a comprehensive collection of quasars with incorrect BALnicity index and redshift values, suggesting a re-evaluation of the BAL_PROB label in the DR16_v4 metadata. The analysis employs Principal Component Analysis (PCA) for spectral decomposition, followed by dual K-Means clustering in a 30-dimensional hyperspace, categorizing the dataset into three clusters. Anomalies are identified based on a 5σ deviation from the cluster centroids and subsequently classified into five distinct groups: Excess SiIV Emitters, Machine Error Anomalies, CIV Peakers, BALs (FeLoBALs& BALs), and Miscellaneous (Group members do not follow a singular unique common trend). A completeness check using CIV, CIII, and MgII flux ratios confirms a 94% true identification rate for CIV Peakers. These anomalies exhibit enhanced and disproportionate spectral features, indicative of unique physical phenomena, thereby facilitating an in-depth understanding of their characteristics. We are currently drafting a manuscript detailing these findings for submission to the Astrophysical Journal by the end of July. We are currently working on developing consistent models to account for the unique behavior of some groups and will also be releasing a detailed catalog highlighting the details of these objects for further analysis by the community.