Paper Title: Proactive Prediction of Network Disconnections Using Machine Learning in Edge Computing
Conference Name: Korean Institute of Communications and Information Sciences (KICS Fall 2024)
Abstract: Network disconnections between edge devices and central servers can lead to significant disruptions in data-driven applications, highlighting a possible need for predictive systems to
anticipate these disconnections. This preliminary study evaluates multiple machine learning models to predict network disconnections k instants ahead using key metrics such as latency,
packet loss, jitter, and congestion levels. This approach involves extensive feature engineering to capture temporal patterns in network performance data, including creating lagged and moving average features. Evaluating the three models yields similar results, with XGboost having the least error. These findings raise interesting questions about the temporal dependence of network metrics and hint at the need for a different approach involving multi-domain data toward forecasting disconnections
in communication networks.