On the Employment of Machine Learning Techniques for Troubleshooting WiFi Networks

The rapidly increasing popularity of 802.11 WLANs along with the co-existence of multiple heterogeneous devices in the unlicensed frequency bands have created unprecedented levels of congestion, especially in densely populated urban areas. Under such complex setups, WLAN under-performance issues experienced by end-users are hard to interpret even by experts. In this paper, we develop an intelligent, easy to deploy mechanism that takes advantage of MAC-layer exported data and employs machine learning techniques to accurately diagnose the five most common WiFi pathologies (contention, low-SNR, non-802.11 Interference, etc.). The collected data are fed to four different classification algorithms, which we fine-tune, in order to optimize their hyper-parameters in regards to their precision and accuracy. The resulting solution provides two different mechanisms, with the first targeting low-overhead passive detection and the second offering more accurate performance relying on active probing. Detection performance is evaluated through extensive testbed experiments and exhibits that the K-Nearest Neighbors classifier achieves almost 100% accuracy and precision for the active probing and 95% accuracy and precision for the passive detection across the five considered pathologies.