The reliability of edge computing platforms becomes important as more technologies such as the Internet of Things ( IoT) are increasingly operating in such an environment. We propose a network topological analysis framework that applies machine learning (ML) to predict network failover and identify the low cost solution. The objectives are multifolds: identify the optimal algorithm(s) to generate network topological features, select the best ML model(s) to predict the hidden linkage and service disruption due to failover of critical nodes, develop an efficient and effective resource reallocation strategy to prevent failover, and evaluate the return on investment. Using continental European power grids dataset, this paper reports the result of applying this topological analysis framework on predicting the most stable flow network flow to prepare for the next phase of preemptive resource reallocation.