Enhancing pH prediction accuracy in AlO gated ISFET using XGBoost regressor and stacking ensemble learning.

An ion-sensitive field-effect transistor (ISFET) is widely used in environmental and biomedical applications due to its rapid response, miniaturization, and cost-effectiveness. In this study, a numerical model of an Al₂O₃-gated ISFET was developed to detect pH levels. The effects of gate dielectric thickness, doping concentration, and temperature on ISFET's performance were evaluated using I-V characteristics. An eXtreme Gradient Boosting (XGBoost) regression model was employed to predict pH levels using data obtained from I-V characteristics. Further, Hyperparameter optimization was performed to tune critical XGBoost-hyperparameters such as maximum depth, minimum child weight, estimators, learning rate, α, and λ. The optimization strategies such as random search, grid search and Bayesian optimization were utilized to improve the efficacy of regressor by minimizing errors and maximizing accuracy in prediction. A stacking ensemble learning approach was also implemented to integrate multiple models, enhancing prediction accuracy and thereby capturing additional information. The XGBoost regressor achieved superior results with R = 0.9846, MSE = 0.2342, and MAE = 0.2317, compared to other regressor models. Therefore, the use of XGBoost regressors with hyperparameter optimization and stacking ensemble learning approach is found to be highly effective for pH prediction from ISFET under various operating conditions.