Application of the Adaptive Neuro-Fuzzy Inference System (ANFIS) and Sobol Approaches for Modeling and Sensitivity Analysis of the Biosorption of Triglyceride from the Blood Serum

An adaptive neuro-fuzzy inference system (ANFIS) was applied to simulate the batch adsorption of triglyceride (TG) from the human blood serum using the cinnamon powder, which has appeared as a potential biosorbent for serum purification, in our previous work. The obtained experimental results were used to train and evaluate the ANFIS model. Temperature (°C), the adsorption time (h), the stirring rate (rpm), the dose of adsorbent (g) and the adsorbent milling time (min) (or the particle sizes of the powder) were considered as the model inputs and TG removal (%) was chosen as the model response. The ANFIS model was trained with 75 % of the available data while 25 % of the remaining data was used to verify the validity of the obtained model. Sobol sensitivity analysis results indicated that the cinnamon dose with 71 % and the adsorbent milling time (or the particle size of the powder) with 15 % impact share were the most influential variables on the TG removal. Furthermore, the specific surface area and the number of reactive adsorption sites were found to be the most important characteristics of the adsorbent. Generally, the results of this study confirmed the advantages of applying the ANFIS and Sobol approaches for the data-based modeling of the bioprocesses.