Bir Wi-Fi ağındaki istasyonların tespit edilmesinde makine öğrenmesi yaklaşımı

Wi-Fi technology has become a vital component of modern life, driving seamless communication and effortless access to information. With its continuous evolution, the number of connected devices has grown exponentially. This rapid expansion has introduced challenges, one of which is station fingerprinting in Wi-Fi networks. In this work, we tackeled this problem, station identification in a Wi-Fi network, using a multi-class supervised classification approach. The dataset is obtained from a real-world scenario from a private ISP, encompasses observations from physical and network layers, traffic patterns and protocol features. Our study integrates these feature groups to develop robust classifiers for identifying ten distinct station types. In this work, we explored various classifiers across 2.4GHz and 5GHz interfaces. A critical focus is on evaluating the performance of these models independently and in combination through stacking methods. Band stacking merges predictions from 2.4GHz and 5GHz of the same classifier type, while an advanced stacking approach integrates outputs of different classifier types. Experimental results compares individual classifier performances against stacking methods. The findings underscore the potential of combining heterogeneous features and leveraging stacked classifiers for station identification, offering a scalable and efficient solution for fingerprinting in a real-world Wi-Fi network set ups.