A trip already appears to be a combination of 2 data: trip=[departure time, arrival time]. What is stopping you from pairing them up with the return trip before the shuffle: daily_commute=[htw_departure, htw_arrival, wth_departure, wth_arrival]? I assume it will be for sampling or bootrstrapping. You could also generate a lookup feature, like date, and have a full datetime format for your data instead of just time.
What do you mean what is stopping me before scuffling? Before shuffling, I can easily pair them up, but my question is that how can I associate the w-t-h trips and h-t-w trips after shuffling.