Feature Compression Based BP Neural Network for IoT Performance Prediction

The widespread access to Internet of Things (IoT) devices has led to an increase in the diversity and uncertainty of wireless networks. This, in turn, presents challenges when it comes to accurately analyzing the performance of wireless networks. In order to study these networks in a flexible and straightforward manner, this paper proposes a wireless network performance prediction method based on a two-stage feature compression Back Propagation (BP) neural network. By employing a multiple input multiple output BP neural network, this method allows for the analysis of the impact of different network parameters/features on network performance metrics. To address the problem of high dimensionality associated with the original network parameters, the proposed method utilizes a two-stage feature compression approach. Firstly, redundant features are merged by leveraging mathematical relationships among the network parameters. Secondly, network parameters are optimally selected based on the correlation between the network parameter data and network performance metrics data, as well as the redundancy among various network parameter data. Simulation results demonstrate that the proposed method is capable of appropriately selecting network features, avoiding feature redundancy, reducing the complexity of BP neural network training, and achieving high accuracy in network performance prediction.