To predict the characteristic impedance of the microstrip transmission line using supervised machine learning regression techniques

Abstract

In this paper, supervised machine learning regression techniques: Support Vector Machine (SVM), Random Forest and Deep Neural Network (DNN) models, are demonstrated to predict the characteristic impedance of the microstrip transmission line. Here, microstrip transmission line width, substrate height and substrate dielectric constant are taken as the input and characteristics impedance as the output parameter. To train the models, the data set is created using microstrip transmission line analytical models. DNN models are developed using Feed-forward Back-propagation learning algorithm, where 'adam' is used as optimiser and 'relu' as the activation function. The regression predictive model of SVM and Random Forest model of ensemble learning using bagging technique are developed. It is found that minimum MSE of DNN model is 0.04191 with high execution time 1114.179655 sec, whereas SVM model shows low execution time of 0.8327 sec with MSE of 0.49. Random Forest model showed the MSE of 0.14 with execution time 1.4296 sec.