A novel hybrid prediction model based outdoor fingerprint localization for internet of things

Cellular network fingerprint localization technology utilizes signal feature analysis for location estimation, providing a crucial technological pathway to enhance the accuracy of location-based services (LBS) in Internet of Things (IoT) applications. However, this technology faces new challenges. Firstly, to address errors in sub-model outputs that traditional hybrid prediction models cannot effectively identify, we designed a novel hybrid prediction model. This model combines a one-dimensional convolutional neural network and a fully connected neural network in parallel, integrating a newly designed filtering process to eliminate most errors in sub-model outputs and reduce subsequent computational burden. Secondly, considering the limitation of fingerprint localization technology in providing localization error information concurrently with predicted locations, we proposed a localization error estimation method. This method offers uncertainty metrics with each predicted location to provide relevant uncertainty measures. Lastly, addressing the current lack of credibility analysis for fingerprint predicted locations, we devised a credibility assessment method aimed at enhancing the reliability of localization results by providing comprehensive information. Carefully selected evaluations in complex urban environments validate the effectiveness of the proposed localization technology. Compared to existing technologies, it demonstrates superior performance with a median localization error of 8.53 m and an average localization error of 13.36 m. In the future, the technology is expected to play a key role in LBS for the IoT, improving the accuracy and reliability of the system.