Lead-free perovskites (A3B2X9) have drawn much attention in recent years. However, a thorough understanding of these materials is still in its early stages. This is because A3B2X9 perovskites have large-scale component tunability, in which the A+, B3+, and X- ions can be replaced or partially substituted with other elements. Here, based on density functional theory and machine learning techniques we propose a data-driven method to find suitable configurations for photocatalytic water splitting. By replacing atoms in A3B2X9, 3.4 million configurations are constructed and studied. Our results show that the substitutional position plays an important role in determining the photocatalytic performance. Specifically, the co-existence of Br and I elements is favorable for X-sites, while for B-site atoms, it is better to choose atoms from groups IIIB and IIIA with a period greater than 3. Considering their rarity and toxicity, we believe that In is a good choice for B-sites and propose CsRb2BiInBr5I4 as a promising candidate. These results may provide guidance for the discovery of novel lead-free perovskites for photocatalytic applications.