Flow timeout matters: Investigating the impact of active and idle timeouts on the performance of machine learning models in detecting security threats