TinyAP: An Intelligent Access Point to Combat Wi-Fi Attacks Using TinyML

In the recent past, there has been a significant surge in MAC layer Wi-Fi attacks, leaving Wireless Local Area Networks (WLANs) vulnerable to different types of network intrusion. The current routers cum Access Point (AP) lack the necessary intelligence to prevent or combat these attacks, thereby making it imperative to rely on various networking intrusion detection systems. To address this issue, we have developed an intelligent AP that can identify different MAC layer attacks and generate alerts as required. This has been achieved using lightweight Machine Learning (ML) models on a microcontroller, leveraging the TensorFlow-lite-micro (TFLite Micro)libraries. A framework called TinyAP has been developed, which uses the efficient Neural Architecture Search (NAS) and the lightweight inference engine to classify Wi-Fi attack inference at the AP level. The system consists of a Raspberry Pi 3 B+, functioning as an AP and an Arduino Nano 33 BLE Sense acting as a plugin. The Nano module executes pre-trained TinyML models, where the models are prepared (i.e., searched, trained and converted) on a Desktop PC (DESKTOP-47M82UC). We employed Neural Architecture Search for Multilevel Perceptrons (MLPNAS) to generate the five best neural architecture models to deploy on the system. To test the efficacy of this innovative approach, an online dataset containing 13 different Wi-Fi-labeled attacks has been used. TinyAP has achieved an average accuracy of 96.45% for all the Wi-Fi attacks and a maximum accuracy of 100% in the 2 Class (2C) classification for some of the Wi-Fi attacks like KRACK and Kr00k. Additionally, TinyAP has also shown a maximum accuracy of 95.19% in multiclass classification. The average accuracy of TinyAP shows a marginal reduction of 0.76% in 2C classification and 0.04% in multi-class classification when compared with Multilayer Perceptron (MLP) models. The promising result of the TinyAP is that it provides a much-needed layer of protection for networks and offers a proactive security solution at the AP level.