Double perovskite oxide A₂BB'O₆ has better stability and wider bandgap range than ABO₃-type oxide, and exhibits great prospects in photocatalytic overall water splitting. However, owing to the diversity of crystal structure and constituents of perovskite oxide, rapidly and accurately searching for A₂BB'O₆ for photocatalyst is still a big challenge, both experimentally and theoretically. In this work, in order to screen out suitable double perovskite oxide photocatalysts, a multi-step framework combined with machine learning technique and first-principles calculations is proposed. Nearly 8000 candidates with proper bandgaps for water splitting are screened out from among more than 50000 A₂BB'O₆-type double perovskite oxides. Statistical analysis of the results shows that double perovskite oxides with d¹⁰ metal ions at B/B' sites are more likely to have good absorption of visible light, and the structural symmetry of double perovskite also has influence on the bandgap value. Furthermore, first-principles calculations demonstrate that Sr₂GaSbO₆, Sr₂InSbO₆ and K₂NbTaO₆ are non-toxic photocatalyst candidates with proper band edges for overall water splitting.