| dc.description.abstract | This paper presents the design and implementation of a microstrip patch antenna-based sensing system for pH estimation of aqueous solutions using machine learning. Conventional pH sensors are widely used, such as frequent calibration, easy contamination, the measurement range is limited, and the maintenance is expensive. In order to overcome these limitations, the proposed system uses rectangular microstrip patch antenna fabricated using FR4 substrate for the detection of change in the dielectric properties of liquid samples. Variations in resonant frequency, return loss (S11) and bandwidth are then analyzed to provide the predict pH level. Experimental measurements were conducted for solutions with a pH value ranging from 4 to
12 by using a Vector Network Analyzer (VNA) with the results verified by simulation results of CST Studio Suite. Furthermore, Random Forest, Support Vector Regression (SVR) and kNearest Neighbors (kNN) machine learning models were also trained with the collected dataset in order to predict the pH with respect to the antenna response parameters. Out of them,
Random Forest showed the highest accuracy (𝑅 2 level of higher than 0.85) with the lowest MAE (Mean Absolute Error) showing also the strong predictive reliability. The obtained results validate the concept that a low-cost, wide-area and contact-less pH monitoring solution is obtained by the combination of microwave sensing and data-driven modeling. The proposed system has potential applications in environmental monitoring, industrial process control, fish farming and agricultural water management especially for resource constrained environments like that of Bangladesh. | en_US |
