Abstract: Exploring the optimum condition for modified spent grains’ adsorption of As(III) is complex, this paper introduces a feature extraction method to help find the influence on adsorption rate for each factor during the experiment, to discover the optimum condition for modified spent grains to adsorb As(III). To extract the relationship between functional group and adsorption properties, both linear principal component analysis method and nonlinear manifold learning method were adopted for dimension reduction. Finally, the quantitative structure activity relationship with support vector regress and partial least squares was established to find out the optimum condition for adsorption. Results show that the modifier concentration is 1mg/L, the PH value of solution is 7, modification time is 12h, reaction temperature is 20℃, adsorption time is 2h, dose of modified spent grain is 0.3mg, and the adsorption rate can reach 96.19%.