A YOLOv8-based travel recommendation system through the analysis of visual data on mobile devices

This study seeks to investigate the processes of real-time object detection and content interpretation on mobile devices utilizing artificial intelligence methodologies, while also designing and developing a Mobile Travel Recommendation System within this framework. The research emphasizes the incorporation of the YOLOv8 object detection algorithm into the Android platform and thoroughly examines its effectiveness on mobile devices. One of the key goals of this research is to allow users to select multiple images from their gallery, analyze them through the Roboflow platform, and offer tailored travel recommendations based on the resulting categorical predictions. The application assesses each uploaded image via the Roboflow API, categorizing them into four primary groups—nature, historical landmarks, summer vacation, and winter vacation—and generates insightful travel suggestions from these classifications. The thesis encompasses not only the fundamental concepts of the YOLOv8 algorithm but also the assessment of training methodologies and deployment efficiency of various model configurations on mobile devices. Consequently, YOLOv8 models trained in a Python environment were integrated with an Android application through API communication. Initially, the study analyzed the structure and operational principles of YOLOv8, comparing it with previous iterations of object detection algorithms. This was succeeded by a comprehensive description of the mobile application development process within the Android ecosystem, model training on Roboflow, API integration, and the architecture of the multi-photo prediction system. During the model training phase, datasets were generated on the Roboflow platform, with images labeled according to their respective categories. The two trained YOLOv8 models achieved accuracy rates of 96.6% and 98.3%, respectively, indicating that effective preprocessing techniques and suitable model configurations significantly enhance overall performance. In summary, the Mobile Travel Recommendation System was effectively designed and implemented as an AI-driven mobile application that can analyze various user-submitted images and provide personalized travel theme recommendations based on individual preferences. This research aims to advance the practical use of AI-driven content analysis and user-focused recommendation systems in the field of mobile technology.