The electrochemical reduction of carbon dioxide (CO2) is a promising approach towards the utilization of CO2 as a carbon feedstock, which is desirable for mitigating global warming and closing the carbon cycle. Despite the desire to develop metal-based electrocatalysts for CO2 reduction, CO2 reduction activity on metal electrodes is limited largely by the scaling relationship of intermediate binding energies. To fabricate efficient electrocatalysts, the adsorption energy of the target intermediate species in each reaction process should be optimized; however, such optimization becomes difficult when intermediate species have similar adsorption properties, resulting in an unavoidable overpotential. Multi-component compounds, such as metal sulfides, have the potential to break this scaling relationship because they possess multiple adsorption sites. Further, the investigation of electrochemical CO2 reduction activity on metal sulfides could provide insights into the origins of life, as metal sulfides have been hypothesized as proto-catalysts of extant CO2 reduction enzymes.Catalyst informatics is an effective approach for predicting the relationship between the physicochemical parameters and catalytic activity of materials. Here, we performed multi-regression analysis of electrochemical CO2 reduction catalyzed by metal sulfides to gain insight into the activity-determining properties of these electrocatalytic materials using following 14 physicochemical parameters as explanatory variables: metal-sulfur bonding length, nearest metal-metal distance, absolute negativity, standard redox potential of metals, band gap, first ionic energy, maximum radius of s-orbital, d-electron number, molar capacity of heat, formation enthalpy, metal-sulfur-metal angle, maximum radius of d-orbital, position of conduction band edge, and electron affinity. To avoid the effect of collinearity among explanatory variables, we applied partial least squares (PLS) analysis for the multi-regression process. PLS extracts the main components and uses the principal component as the explanatory variable. As the correlation between the main components is set to zero, collinearity can be avoided. Further, the number of main components can be controlled, allowing for regression to be performed with fewer samples. In addition, we applied the least absolute shrinkage and selection operator (LASSO) regression analysis.To obtain objective variable for regression analysis, the synthesized metal sulfides were subjected to electrochemical CO2 reduction. After applying an electrochemical potential of -1.2 V vs. SHE for 1 h, CO and HCOO- were obtained as CO2 reduction products for several of the metal sulfide samples. Among the 14 samples examined (Ag2S, CdS, CoS2, CuS, Fe3S4, In2S3, MnS, MoS2, NiS2, Ni3S2, SnS2, TiS2, WS2, and ZnS), CdS and Ag2S exhibited the highest CO production activity, with Faradaic efficiencies of 37.2% and 29.5%, respectively, in a 0.1 M KHCO3 aqueous solution, whereas CuS maximized formate (HCOO-) production, with a FE of 4.5%. PLS and Lasso analyses revealed that CO production was largely affected by the structural parameter of metal-sulfur bond length; however, HCOO- generation was influenced by the electronic parameters of the metal centers, particularly absolute electronegativity. These findings are distinct from those of metal electrodes, for which the generation of CO and HCOO- is predominantly influenced by electronic parameters, demonstrating that CO2 reduction on metal sulfides proceeds by a different mechanism compared to that of the corresponding metal electrodes. As demonstrated in the present study, changing the structural parameters, such as metal-sulfur bond length, alters the adsorption characteristics of the intermediate species to optimize the CO/HCOO- production activity. We anticipate that the key parameters identified through multi-regression analysis of metal sulfides will prompt further study of the factors influencing the specificity of these unique electrochemical CO2 reduction catalysts.