Machine learning approach to optimization of parameters for impedance measurements of Quartz Crystal Microbalance to improve limit of detection

Impedance spectroscopy allows real time monitoring of fundamental resonance frequency and harmonics of AT cut Quartz Crystal Microbalance (QCM) biosensors. Acquisition of accurate and high resolution spectroscopic data requires optimization of parameters, which is rarely performed as it requires an excessive number of experiments. In this work, machine learning clustering and classification algorithms are applied in an attempt to minimize the number of experiments by designating a small representative subset of chosen values of parameters, while preserving the variation of those values. Our results indicated it is possible to reduce the number of experiments required by more than 10 fold. The constructed model has both an unsupervised clustering part (k-means algorithm) and a supervised classification part (Support Vector Machine algorithm), working independently. From those two different approaches 83% compatibility is obtained. Such optimization can improve limit of detection (LOD) of glycerol concentrations by a factor of 12 using a QCM with a fundamental resonance frequency of 10 MHz. This approach can be applied not only to biosensors but also to chemical sensors involving resonance frequency monitoring by means of impedance spectroscopy, and has the potential to be expanded to systems with different transducer types, monitoring resonance by other means.