Enhancing the Predictive Performance of Molecularly Imprinted Polymer-Based Electrochemical Sensors Using a Stacking...

Area of Science:

• Electrochemistry
• Materials Science
• Data Science

Background:

• Electrochemical sensor performance relies on balancing multiple influencing factors.
• Machine learning (ML) offers powerful tools for analyzing sensor parameters and predicting performance.
• Developing selective and sensitive sensors, like those for doxorubicin (Dox), requires advanced modeling techniques.

Purpose of the Study:

• To develop a molecularly imprinted polymer (MIP)-based electrochemical sensor for doxorubicin (Dox) detection.
• To enhance sensor performance and reliability using ML-based ensemble models.
• To investigate feature importance and optimize model predictive capabilities.

Main Methods:

• Utilized four ML models (DT, XGBoost, RF, KNN) with SHAP for feature importance analysis.
• Applied min-max scaling to ensure proportional feature contribution.
• Compared performance of multiple ML models (LR, KNN, DT, RF, AdaBoost, GB, SVR, XGBoost, Bagging, PLS, Ridge) for predicting sensor output current.
• Developed a novel stacking regressor ensemble model integrating DT, RF, GB, XGBoost, and Bagging.

Main Results:

• Feature importance analysis guided optimization by removing less influential and adding new features.
• The proposed stacking regressor ensemble model significantly outperformed individual ML models.
• Achieved a high R-squared (R²) of 0.993, with a reduced RMSE of 0.436 and MAE of 0.244.
• Demonstrated enhanced sensitivity and reliability of the MIP-based electrochemical sensor.

Conclusions:

• The developed stacking regressor ensemble model offers a robust method for improving MIP-based electrochemical sensor performance.
• This ML-driven approach enhances prediction accuracy and reliability for Dox detection.
• The methodology is adaptable for developing other electrochemical sensors with diverse transducers and sensing elements.