Enhancing random forest model prediction of gas holdup in internal draft airlift loop contactors with genetic algorithms tuning and interpretability

Gas holdup in an Internal Loop Airlift Contactor (ILAC) is one of the primary hydrodynamic features that governs the performance of the contactor. Prediction of gas holdup assumes importance in the design of airlift contactor. The key challenge in prediction of gas holdup is due to the difficulty in obtaining reliable phenomenological models. The improved prediction accuracies can provide better contactor design and equipment performance. The current work focuses on the use of data driven models for prediction of gas holdup, as the existing empirical correlations were found to be inadequate. In this work, 324 data points from the reported works on internal draft airlift contactor are consolidated. The input part of the examples in the data set consists of ten features which are related to geometry, fluid properties and operating conditions. This study improves the predictive accuracy of gas holdup, the output variable, in ILACs using a Random Forest (RF) model optimized via Genetic Algorithms (GA). SHapley Additive exPlanations (SHAP) based interpretability reveals the contribution of key features. The tuned model achieved a Coefficient of Determination (R²) score of 0.9542 and Mean Absolute Error (MAE) 0.0059, surpassing traditional parameter sets and providing insights into hydrodynamic control.