Abstarct: In the first section of this research, the seed husk of Fraxinus syriaca modified with NaOH was used to absorb crystal violet (CV) and methylene blue (MB) dyes in binary system. The prepared adsorbent was characterized by field emission scanning electron microscopee (FESEM), Fourier transform-infrared (FT-IR), Brunauer–Emmett–Teller (BET) specific surface area and energy dispersive X-ray (EDX). Because of spectral overlapping the partial least square (PLS) modeling baxsed on singular value decomposition (SVD) was used for the multivariate calibration of the spectrophotometric data to simultaneous determination of dye concentration. Screening of most important factors using Plackett-Burman design, used to select the most statistically significant parameters. Optimization of effective experimental parameters such as pH, initial dye concentration, adsorbent dosage and investigation of any possible interaction between variables, carried out using Face centered central composite design (FCCCD) under response surface methodology (RSM) baxsed on adsorption efficiency. According to the results obtained from desirability function, optimum pH, initial dye concentration of each dye and absorbent dosage found 7.6, 30.0 mg/L and 0.380 gr/L respectively in (100.0 ml dye aqueous solution, temperature of 298°K and 20 min of contact time). Kinetic and isotherm studies on data gained from equilibrium condition for different concentration of dyes were showed better fitted in pseudo-second-order (PSO) and extended Langmuir models and consequently the maximum adsorption capacity of CV and MB in their binary component system were found 191.60 and 228.00 respectively. Determination of thermodynamic parameters according to the negative value of ΔG and positive values of ΔH and ΔS confirmed that the adsorption process is spontaneity, endothermic and irregularity increases at the solid-liquid interface. In the second section of this study, in order to generate linear and non-linear models such as the multiple linear regression (MLR), smoothly clipped absolute deviation (SCAD), random forest (RF), and artificial neural network (ANN), to predict the removal efficiency of these dye species at different experimental conditions, considering empirical parameters such as pH, adsorbent dosage, initial dye concentration and contact time as inputs, and removal efficiency as output of these models. The validation studies of these models were performed using the test set (15% of data) the results obtained showed that the ANN model (as a non-linear model) had a higher ability to predict the adsorption process for both dyes under different experimental conditions.