Prediction of signal attenuation due to air pollutants using bayes-DCNN

Abstract In modern wireless communication, radio waves are crucial for signal transmission between mobile devices, but they are subject to attenuation when interacting with air pollutants such as PM10 particles. This research proposes a novel Bayes-Deep Convolutional Neural Network (Bayes-DCNN) to predict signal attenuation caused by air pollution. The model incorporates Bayes by Backprop to manage uncertainty and improve training accuracy. Experiments demonstrate that the proposed Bayes-DCNN outperforms existing methods, achieving an accuracy of 95.52%, compared to 92.15% for CNN, 91.11% for SVM, and 89.47% for LSTM. The model also exhibited superior precision (94.38%), recall (93.50%), and F1-score (94.36%) when evaluated against other deep learning and non-deep learning approaches. Additionally, Bayes-DCNN reduced root mean squared error (RMSE) to 0.143, outperforming CNN (RMSE of 0.146), SVM (RMSE of 0.155), and LSTM (RMSE of 0.167). These improvements demonstrate the efficacy of the proposed model in predicting signal attenuation under various air pollution conditions.