Random Forest Assisted RF Fingerprinting Positioning for Cellular Network

Contextualization is a core of 6th Generation Mobile Communication Technology (6G), and communication perception has gradually become one of the critical business application scenarios. Therefore, it becomes particularly important to refine the classification of various intelligent scenarios and provide personalized services for mobile users. However, the constantly changing of scenarios is still a challenge for current network operators, which indicates that accurately identify and predict user location is the key to location-based services (LBS). In this paper, a random forest assisted radio frequency (RF) fingerprinting positioning method based on cellular big data is proposed to eliminate the interference of indoor fingerprints with incorrect GPS information on outdoor positioning. Firstly, we collect a large number of measurement reports (MRs) by minimized drive test (MDT) from typical urban areas covered by the present 4G Long Term Evolution (LTE) cellular network. Then, Indoor-Outdoor (IO) classification is realized using random forest after conducting feature engineering. Finally, RF fingerprinting positioning is performed on the basis of classification, and it can be found that the positioning error of the outdoor MRs based on purification was effectively reduced compared to the initial data. Specifically, when the probability is 67%, the positioning error after IO classification is about 5% lower than the original base.