Abstract:
Sri Lankan tourism is rapidly evolving, creating growing demand for intelligent tour planning
systems that offer optimized and personalized travel experiences. Existing platforms, however,
are inflexible, lack personalized routing, and do not include lesser-known destinations limiting
user engagement and reducing the overall tourism experience.
This paper describes Serendib Yathra, an intelligent, machine learning-driven tour planning
system that can generate user-specific travel plans based on user preference and points of
interest. Unlike static tour planners, Serendib Yathra employs a hybrid machine learning
approach consisting of TF-IDF-based content filtering and K-Means clustering for proposing
suitable points of interest (POIs) and graph-based route optimization (utilizing Dijkstra's
algorithm) to construct efficient travel routes via different cities.
The system is not reliant on real-time APIs or dynamic user input and operates exclusively on
preprocessed static data sets like user profiles, travel time, and filtered POI data. The platform
was developed with Python and Flask via an Agile development cycle with iterative user testing
and tuning.
Final evaluation reflected more than 80% success in suggesting individualized attractions
related to the users' interest. User tests also reflected a high level of satisfaction with routing
efficiency and quality of itinerary in general. The system has excellent prospects in ensuring
sustainable tourism through support for popular and less seen sights, as well as for more
overall personalization of the trip planning process within Sri Lanka.
