Abstract
This study provides an in-depth examination of forecasting the concentration of pharmaceutical compounds utilizing the input features (coordinates) r and z through a range of machine learning models. Purification of pharmaceuticals via vacuum membrane distillation process was carried out and the model was developed for prediction of separation efficiency based on hybrid approach. Dataset was collected from mass transfer analysis of process to obtain concentration distribution in the feed side of membrane distillation and used it for machine learning models. The dataset has undergone preprocessing, which includes outlier detection using the Isolation Forest algorithm. Three regression models were used including polynomial regression (PR), k-nearest neighbors (KNN), and Tweedie regression (TWR). These models were further enhanced using the Bagging ensemble technique to improve prediction accuracy and reduce variance. Hyper-parameter optimization was conducted using the Multi-Verse Optimizer algorithm, which draws inspiration from cosmological concepts. The Bagging-KNN model had the highest predictive accuracy (R2 = 0.99923) on the test set, indicating exceptional precision. The Bagging-PR model displayed satisfactory performance, with a slightly reduced level of accuracy. In contrast, the Bagging-TWR model showcased the least accuracy among the three models. This research illustrates the effectiveness of incorporating bagging and advanced optimization methods for precise and dependable predictive modeling in complex datasets.
