Integrated Corridor Management (ICM) has the potential to provide significant benefits in managing the congestion at the corridor level. One of the crucial ICM strategies to facilitate the diversion of traffic to underutilized route(s) from the impacted route(s) during incidents. Identifying alternative routes to the impacted routes and estimating the diversion rates will allow the agency to develop and deploys new management and signal control plans on the alternative route(s). To accomplish this, there is a need to identify the critical alternative routes among the available routes for the specific type, time of day, and location of the incidents. This is important since diversion to alternative routes is a function of incident and traffic attributes and change dynamically based on the attribute values. This study introduces the use of a deep learning approach based solution to assist the agency in identifying the alternative routes used by motorists. The long short term memory (LSTM) deep neural network method is applied to develop a model capable of predicting the travel time in a selected horizon utilizing both incident and traffic attributes. The predicted travel time obtained from the model is used to estimate the percentage changes in the travel time (Δ-Travel Time) between incident period and normal period in the all potential alternative routes for different time horizons after the incident. The Δ-Travel Time acts as a threshold measure for the agency to identify the critical route(s) and critical timeframe(s) for implementing the special plans. The developed method is based on travel time data that requires small resources to obtain and may already be available to the agency but provide very high fidelity in identifying critical routes and critical timeframes.