We identify multi-wavelength counterparts to 1147 submillimeter sourcesfrom the S2COSMOS SCUBA-2 survey of the COSMOS field by employing arecently developed radio+machine-learning method trained on a largesample of Atacama Large Millimeter/submillimeter Array(ALMA)–identified submillimeter galaxies (SMGs), including 260SMGs identified in the AS2COSMOS pilot survey. In total, we identify1222 optical/near-infrared (NIR)/radio counterparts to the 897 S2COSMOSsubmillimeter sources with S 850 > 1.6 mJy, yielding anoverall identification rate of (78 ± 9)%. We find that (22± 5)% of S2COSMOS sources have multiple identified counterparts.We estimate that roughly 27% of these multiple counterparts within thesame SCUBA-2 error circles very likely arise from physically associatedgalaxies rather than line-of-sight projections by chance. Thephotometric redshift of our radio+machine-learning-identified SMGsranges from z = 0.2 to 5.7 and peaks at z = 2.3 ± 0.1. The AGNfraction of our sample is (19 ± 4)%, which is consistent withthat of ALMA SMGs in the literature. Comparing with radio/NIR-detectedfield galaxy population in the COSMOS field, ourradio+machine-learning-identified counterparts of SMGs have the higheststar formation rates and stellar masses. These characteristics suggestthat our identified counterparts of S2COSMOS sources are arepresentative sample of SMGs at z ≲ 3. We employ ourmachine-learning technique to the whole COSMOS field and identified 6877potential SMGs, most of which are expected to have submillimeteremission fainter than the confusion limit of our S2COSMOS surveys({S}850μ {{m}}≲ 1.5 mJy). We study the clusteringproperties of SMGs based on this statistically large sample, findingthat they reside in high-mass dark matter halos ((1.2 ± 0.3)× 1013 h ‑1 {M}ȯ ),which suggests that SMGs may be the progenitors of massive ellipticalswe see in the local universe.