Non-WiFi Interference Detection and Throughput Estimation at the WiFi Edge for 2.4 and 5 GHz Bands with Machine Learning

Non-WiFi interference at the WiFi Edge for the 2.4 and 5 GHz bands is investigated. We propose machine learning models for detecting the non-WiFi interference and estimating the throughput based on WiFi quality of service statistics. Non-WiFi interference sources utilized in the tests are a residential-type microwave oven, a baby monitor, and a media converter. The proposed solutions do not require specialized hardware, chipset, or end-user engagement; they operate only on WiFi statistics collected from commercial access points. We implement decision tree (DT), random forest (RF), CatBoost (CAT), and multilayer perceptron (MLP) algorithms for both applications. For throughput estimation, an analytical metric is also implemented as a benchmark to compare the performances of the proposed models. RF models outperformed the other proposed models and the benchmark metric.