TRAVEL RECOMMENDATION SYSTEM

Abstract: This paper proposes and experimentally validates a production-ready hybrid machine learning framework for intelligent travel destination recommendation, combining collaborative filtering and content-aware modelling within an integrated Flask-based deployment environment. We construct a consolidated dataset of 1,257,000 anonymized user–destination interactions aggregated from multiple open travel datasets and simulated user profiles. The system employs a modular feature-engineering pipeline that extends a 12-dimensional raw feature space (user demographics, ratings, destination tags) into 210 engineered descriptors incorporating semantic embeddings, recency-weighted interaction scores, and geo-temporal correlations.Training was performed with stratified 5-fold cross-validation and temporal validation splits to prevent leakage. On the reserved evaluation set, the proposed model achieved Precision@10 = 0.314, Recall@10 = 0.283, MAP@10 = 0.301, and RMSE = 0.925, outperforming baseline collaborative filtering (Precision@10 = 0.211). Flask-based deployment yielded mean response latency of 78 ms per query under concurrent load, confirming suitability for real-time applications. Ablation studies revealed the largest marginal gain from the semantic-content embedding layer, enhancing personalization for cold-start users. The system requires no proprietary data and is deployable on commodity hardware, providing a reproducible and scalable baseline for academic and industrial tourism analytics. Future directions include integration of context-aware deep models, federated personalization, and reinforcement-based travel itinerary optimization.

Keywords: Hybrid Machine Learning, Collaborative Filtering, Intelligent Travel Recommendation, Matrix Factorization.