Predicting Tourists' Accommodation Location Scores Using Spatial Machine Learning Techniques A Case Study of Middle Vancouver Island

This study develops a predictive framework to optimize site selection for tourist accommodations - including hotels, motels, resorts, and guest houses (HMRG) - across the central and northern regions of Vancouver Island, aiming to reduce investor uncertainty through data-driven decision support. Unlike traditional models that focus on price prediction, this research emphasizes predicting location scores, a less explored yet highly relevant metric for assessing accommodation desirability. Despite a relatively small sample size, the framework offers promising insights for early-stage modeling in emerging markets. By integrating geospatial analytics and customer sentiment data, the study evaluates three techniques - Ordinary Least Squares Regression (OLSR), Random Forest (RF) regression, and Multilayer Perceptron (MLP) regression - to identify key determinants of location suitability. A four-phase methodology was employed: (1) variable selection and preprocessing, prioritizing tourism-relevant spatial features extracted from user-generated content and refined through spatial data engineering; (2) evaluation of predictor effect sizes, directional relationships, and multicollinearity; (3) iterative model optimization through feature engineering and hyperparameter tuning; and (4) comparative validation using robustness metrics.