A Multi-Level Genetic Algorithm Approach for Generating Efficient Travel Plans

Travel planning is a challenging combinatorial problem that requires automated computation. Given a number of destinations to be visited by a traveler, his/her accommodation location, the duration of his/her visit, and some constraints for each destination, a travel plan application is expected to create a schedule for visiting the chosen destinations. Determining such a schedule manually is a tedious and time-consuming task due to the algorithm complexity which falls in O(n!) if the problem is to be solved by using brute force approach. In this research, the problem is treated as a traveling salesman problem (TSP) and solved using genetic algorithm (GA), which has been widely known to be capable of solving combinatorial problems. However, to employ GA in travel planning, there are challenges in determining the appropriate fitness model and its parameters. To this end, this research performs two experiment scenarios. The first scenario is to compare multi-level GA to single-level GA in solving the problem and to define the optimal parameters. We found that the multi-level GA obtains 108 minutes less trip duration. The second scenario is to evaluate the solution found by the multi-level GA. The trip duration is very close to the solution of brute force approach, with only 1 minute different, but with significantly faster processing time by 50 seconds compared to 20 minutes. These results confirm that our multi-level GA implementation is proven to be applicable for the problem of generating travel plan.