Efficient 5G MIMO antenna development using machine learning-driven parameter prediction

This paper investigates machine learning algorithms (ML) to improve 5 g antenna. The proposed antenna consists of incorporates a combination of circular and semi-circular slot-loaded elements arranged in a 2 × 4 configuration and placed on a substrate with dimension of 30.725 × 30.725 mm². the proposed multi input multi output (MIMO) antenna, fabricated on Rogers RT/duroid 5880 substrate. To enhance the performance of the antenna in terms of bandwidth and gain by using ML guided approach, which Fourier Series Regression (FSR), Sum Sine Regression (SSR), and Gaussian Process Regression (GPR) were the three main regression models used for antenna optimization, to see how well the algorithms work, we are using statistical measures like R² (Coefficient of Determination), RMSE (Root Mean Squared Error), and SSE (Sum of Squared Errors). The data includes key geometrical parameters of the antenna structure . The S11 parameter, which we are trying to predict, is in the seventh column of the data. We split the data into 80% for training the model and 20% for testing it. The results showed a clear difference in performance. where the FSR model obtained R²=0.9444, RMSE=0.9801, and SSE=108.5365. As well the SSR model achieved R²=0.9131, RMSE=1.2593, and SSE=169.6838. So, the GPR model performed significantly better, achieving a much higher R² of 0.9886 and a much lower RMSE of 0.4374. Then compared to the FSR and SSR models, GPR's estimates for S11 were closer to the actual measured value. And for the gain, it appears that the Sum of Sine Regression (SSR) model provides the best overall result with highest Coefficient of Determination (R² = 0.92087) and the lowest errors (RMSE = 0.49659, SSE = 154.62), the Fourier Series Regression (FSR) model is also strong and close (R² =0.91153), while the Gaussian Process Regression (GPR) model ranks slightly lower for this parameter as it remains accurate (R² = 0.90617). the study shows that Gaussian Process Regression (GPR) gives the most correct guess for the S11 parameter, getting an R² of 0.9886. Though, for antenna gain, the Sum of Sine Regression (SSR) model was better, giving the highest R² of 0.92087. These results show that GPR is the best choice for return loss, while SSR is best for gain in this antenna design. This ML way points to the use of certain regression models for different antenna measures.