Muhammad Isa Ansori, Wiwik Anggraeni, Retno Aulia Vinarti, Arwin Datumaya Wahyudi Sumari
Address reconstruction in regions without standardized gazetteers presents a significant challenge for geocoding systems and for last-mile delivery operations. In Bali, Indonesia, residential locations often adhere to Banjar-based naming conventions rather than formal street identifiers, leading to spatial ambiguity and reduced accuracy in automated routing. This study introduces a gazetteer-free hybrid framework that combines multimetric fuzzy matching with a Hierarchical Deep Q-Network (HRL-DQN) to predict non-standardized addresses. The fuzzy module integrates the Levenshtein Distance, Partial Ratio, and Token Sort Ratio to produce a Hybrid Mix Score that stabilizes linguistic variability in informal address text. This structured similarity representation serves as an input to the HRL-DQN architecture, which models the hierarchical spatial dependencies across the district, village, and Banjar levels. The framework was trained and evaluated using 17,354 cleaned operational delivery data. The experimental results indicate significant improvements in both spatial coherence and operational efficiency. The mean spatial deviation decreased from 3,385 m under manual workflows to 879 m with the hybrid framework, with case-level spatial dispersion contracting from multi-kilometer spreads to sub-kilometer clusters. Furthermore, the total address handling time was reduced from 12.78 min to 0.292 min per shipment, representing a 97.7% reduction. Comparative analysis confirmed that hierarchical decomposition and reward-driven spatial alignment are the primary contributors to performance gains beyond fuzzy similarity aggregation alone. The proposed framework establishes a scalable methodological foundation for intelligent address reconstruction in informal, non-gazetteer environments and offers practical applicability to culturally embedded addressing systems in developing regions. © 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Department of Information Systems, Institut Teknologi Sepuluh Nopember (ITS), Surabaya, Indonesia; Department of Information Technology, State Polytechnic of Malang, East Java, Malang, 65141, Indonesia