Data-Driven Methods for Wi-Fi Anomaly Detection

This paper presents a methodology based on data science techniques and ML for detecting anomalies in Wi-Fi networks. The approach employs dynamic time series analysis to identify anomalies by monitoring fluctuations in signal strength, packet loss, and other network metrics. SARIMA, SARIMAX, VARIMAX, and LSTM models were applied to address the challenge of forecasting anomalies. Subsequently, a SVM classifier is utilized to identify affected routers at the detected timestamps. Furthermore, the paper explores a technique to enhance the model's adaptability to different router types, considering potential variations in expected performance. This technique, constructed via PCA, enhances the methodology's robustness across diverse network environments, and facilitates its application in various settings. In addition, the paper takes initial efforts to comprehend and define anomalies in a broader sense and specifically explores the correlation with Wi-Fi metrics. This recurrent theme persists throughout the entirety of the paper.