Computational Intelligence and K-Nearest Neighbour-Based Approach to Predict Zn(ii) Ions Biosorption from Blood Plasma-Protein; Central Composite Design and Machine Learning Strategies

Central composite design (CCD) and machine learning (ML) strategies through the aid of response surface methodology (RSM) and k-nearest neighbor (KNN) were employed to forecast the in-vitro decontamination of Zn(ii) from blood plasma-protein (BPP) using a decontaminating agent, Opuntia fragalis leaf (OFL). The blood plasma-protein (BBP) was characterised by the presence of Zn(ii). After that, spiking with known concentrations of Zn(ii) ions was performed to ensure optimal decontamination efficacy of the biosorbent (OFL). This study employed three responses and generated models to capture the simultaneous interactive effects of the independent process factors as they influence the responses based on the residual Zn(ii) ions concentration in the BPP (Q1), the concentration of Zn(ii) ion decontaminated by OFL biosorbent-removal efficiency (Q2), and OFL biosorbent recovery efficacy-desorption (Q3). The functionality of the KNN models was contrasted with RSM models using the full dataset (KNN1) and partitioned data (KNN2) criteria, with the correlation coefficient (R2) and root mean square error (RMSE) as metrics. Parameter tuning was performed to optimize the performance of the developed KNN models. It was found to be significantly influenced by the nearest neighbour’s k-parameter, attributed to the disparity in the two approaches. The KNN1 model showcased better performances characterised by higher R2 = 0.8190 - 0.9985 and lower RMSE = 0.1052 - 2.1291 against the RSM model of R2 = 0.7418-0.9564 and RMSE = 0.9105 - 1.250. As per the KNN2 models, although the performance was lower, the decontamination efficiency was higher than that of the RSM models.