Abstract

This study analyzes infrastructure accessibility patterns in Pekalongan City using a grid-based approach and machine learning methods. By integrating data from BPS, OpenStreetMap, and ESRI 2023, the research employs 100m × 100m grid analysis units to measure accessibility to public facilities such as education, healthcare, and commerce. Analysis using three clustering methods (K-Means, Bisecting K-Means, and Agglomerative) identifies three distinctive accessibility patterns. The first cluster (40.29%) demonstrates optimal accessibility with high road density, predominantly in the city center. The second cluster (31.64%) exhibits moderate accessibility, characterizing transitional areas. The third cluster (32.90%) shows the lowest accessibility, particularly in southern and coastal regions. Machine learning modeling using Catboost achieves the highest accuracy with a logloss value of 0.0091, confirming distance to healthcare and commercial facilities as key determinants of accessibility. These findings provide empirical foundations for more targeted infrastructure development, with policy recommendations tailored to each cluster's characteristics. The developed methodology offers a novel approach to urban accessibility analysis that can be replicated in other cities with similar characteristics.