Multi-Ion-Sensing Emulator and Multivariate Calibration Optimization by Machine Learning Models

One paramount challenge in multi-ion-sensing arises from ion interference that degrades the accuracy of sensor calibration. Machine learning models are here proposed to optimize such multivariate However, acquisition big experimental data is time and resource consuming practice, necessitating new paradigms efficient for these data-limited frameworks. Therefore, a novel approach presented this work, where emulator designed explain response an ion-sensing array mixed-ion environment. A case study performed emulating concurrent monitoring sodium, potassium, lithium, lead ions, medium representative sweat samples. These analytes relevant examples applications physiology, therapeutic drug monitoring, heavy metal contamination. It demonstrated calibration datasets output by accurately polymeric solid-contact ion-selective electrodes, root-mean-squared error 1.37, 1.44, 1.78, 2 mV obtained, respectively, Na + , K Li Pb xmlns:xlink="http://www.w3.org/1999/xlink">2+ artificial sweat. Besides, synthetic custom size generated train, validate, evaluate different types regressors. Multi-Output Support Vector Regressor (M-SVR) as compact, accurate, robust, model. features 13.22% normalized root mean squares, 20.29% squares improvement compared simple linear regression unbiased estimator large datasets, its average generalization 3.22%. M-SVR have lower computational complexity than single-output SVR or neural network models, making them suitable solution memory energy-constrained edge devices used continuous real-time multi-ion monitoring.