Özet:

Transmitter polymer is used in many technological applications, such as composite materials, supercaptors, varistors or sensors. It is very important for technological applications that these materials demonstrate a non-linear behavior depending on the basic parameters such as the electronic characteristics, the contribution rate and temperature, such as dielectrous passiveness. Due to the excess of possible parameter configurations, it is difficult and expensive to experimentally identify the sets of all appropriate parameters in order to provide the desired values of these features. It is very important to recommend a practical method for this purpose. In this study, three different machine learning algorithms were applied to estimate the real ( ε' ) and virtual ( ε' ) components of complex dielectrial functions of CM1 and KYZ13 polymer composites contributed basalt and the non-linear changes related to angle frequency (w) and PANI concentration (wt) (10%, 25% and 50%). These algorithms are rectal regression (LR), k- nearest neighbor regression (k-NN), and decision tree regression (DT-R) machine learning algorithms. The set of data required for the training processes of these algorithms is created using experimental dielectrial measurements. These experimental data were obtained using an empedance analyzer in a frequency range ranging from 100 Hz to 17.5 MHz at room temperature. In this study, the predictive performance of linear regression, k-the closest neighboring regression and decision tree regression algorithms was compared in detail. The comparison is determined by the comparison of statistical performance metrics, statistical fit, determination rate (R^2), average square error (OKH) and average absolute error (OMH). Although the linear regression algorithm has not shown a proper and stable predictive performance, the decision tree regression model has been shown to be a more suitable model in the calculation of both real (ε^) and virtual components (ε^).

Anahtar Kelimeler: