Predicting Average Daily Rate for 4-5 Star Hotels and Resorts in Thailand

A Comparative Machine Learning Approach Inte-grating Physical Attributes and Site Characteristics

Abstract

Thailand's reputation as a global tourism destination is underpinned by its diverse natural attractions and a rapidly expanding hospitality sector. Online Travel Agencies (OTAs) have become the primary intermediary between hotels and tourists, generating large datasets that require advanced analytical tools to extract pricing insights. This study applies a comparative machine learning framework within Altair AI Studio to decode the complex, nonlinear determinants of hotel Average Daily Rate (ADR) in Thailand, moving beyond traditional linear models. A dataset of 500 four and five-star hotels was curated from Agoda. Five modeling techniques were systematically evaluated: Multiple Linear Regression, Decision Tree, Gradient Boosting, Deep Learning, and Random Forest. Optimized via 10-fold cross-validation, Random Forest emerged as the best-performing model, achieving an R2 of 0.786 and an RMSE of 33.99 USD. Global feature importance analysis identified Premium Market Tier, Star Rating, Room Size, Site Character Beach, and Facilities Total as the primary determinants of hotel pricing. To enhance model interpretability, LIME and multi-dimensional perturbation analyses were applied to stress test the local network and examine threshold behaviors of high-influence features. The empirical findings were further validated through a structured expert panel comprising eight executive industry professionals, confirming high qualitative scores for research reliability (3.830) and practical business applicability (3.910). This study provides a rigorous, data-driven framework for hotel developers, investors, architects, and revenue managers to optimize pricing strategies and spatial resource allocation.