Machine learning based on patch antenna design and optimization for 5 G applications at 28GHz

• Two proposed patch antenna design at 28GHz. • To achieve high efficiency, good return loss; improve bandwidth, and IEEE standard VSWR for proposed antenna design. • Machine learning-based predictive model and ANOVA test was developed to valid the design process. This study has meticulously designed, tested, and optimized a patch antenna with a rectangular shape at 28 GHz. The antenna utilizes two different substrate materials, each having a significantly different relative permittivity. The two substrate materials used are Rogers RT5880 for Design-I and FR-4 for Design-II. The data reveal a notable return loss, gain, directivity, efficiency, and bandwidth change due to the substrate materials' differing relative permittivity and thickness values. Design I achieved return loss, VSWR of -59.289 dB, 1.0023 and Design II are return loss, VSWR of -49.182 dB, 1.007. Design I achieved gain and directivity of 7.63 dBi and 8.51 dBi, and Design II is 3.98 dBi and 7.56 dBi, respectively. The efficiency of the design-I was 89.66%, and design-II was, 52.65%. In this paper, a predictive model was developed using a polynomial regression algorithm to create a predictive model for the designed antenna. The predictive model accelerates the design process and eliminates the need for rigorous physical prototyping and testing. This study is unique in that it creates a mathematical model using the Multivariate Polynomial algorithm for all three parameters: bandwidth, Return loss, and VSWR. A MPR was also deployed to build mathematical models on the antenna parameters. The evaluation of the predicted outcomes shows that the MPR model achieved an R 2 score of 0.9999, 0.9968, 0.9985 and RMSE of 0.0, 0.0, and 0.25; respectively. The study utilized analysis of variance (ANOVA) to examine the impact of different independent factors on both responses and model validation.