Evaluating Matrix Factorization Techniques for Thematic Mapping of Wilderness Walkability Using Multiple GPX Datasets

Abstract. Quantitative thematic mapping of walkability in wilderness areas is challenging due to sparse and unreliable data. Unlike urban walkability, which depends on built infrastructure, wilderness walkability is influenced by natural terrain features such as slope, surface stability, and vegetation density. This study leverages 1,620 GPX trail datasets from Croatia to infer walkability by analyzing movement speed across spatial cells. To extract latent walkability patterns, we apply matrix factorization techniques, including Singular Value Decomposition (SVD), Non-Negative Matrix Factorization (NMF), Stochastic Gradient Descent (SGD), Alternating Least Squares (ALS), and Fast Independent Component Analysis (FastICA). Evaluation results indicate that NMF and Truncated SVD yield the most accurate and interpretable walkability maps. These findings highlight the potential of matrix factorization for mapping hidden variables in geospatial studies and suggest applications in related fields such as fire risk assessment.