Ansori, Muhammad Isa (2026) Pengembangan Sistem Berbasis Pengetahuan Untuk Prediksi Lokasi Alamat Rumah Masyarakat Bali Tanpa Menggunakan Gazetteers Pada Tahap Akhir Pengiriman Barang. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7026231001-Doktoral.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7026231001-Doktoral.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy |
Abstract
Keberhasilan pengiriman barang pada tahap akhir (last-mile delivery) sangat ditentukan oleh ketepatan identifikasi lokasi alamat rumah. Di Bali, banyak alamat ditulis berbasis Banjar yang tidak tercatat dalam basis data geografis formal, sehingga pendekatan geocoding berbasis gazetteers kurang efektif. Penelitian ini mengembangkan sistem berbasis pengetahuan untuk memprediksi lokasi alamat rumah masyarakat Bali tanpa gazetteers dengan mengintegrasikan multi-metric fuzzy matching dan Hierarchical Reinforcement Learning berbasis Deep Q-Network (HRL-DQN). Fuzzy matching menggabungkan Levenshtein Distance, Partial Ratio, dan Token Sort Ratio melalui Hybrid Best-Score (Mix Score 0–100) untuk merepresentasikan kedekatan linguistik alamat tidak terstruktur, sedangkan HRL-DQN memodelkan prediksi hierarkis pada tingkat desa/kelurahan dan Banjar. Evaluasi menggunakan data operasional pengiriman Pos Indonesia di Kabupaten Gianyar sebanyak 17.456 data histori kiriman menunjukkan bahwa alur kerja usulan meningkatkan efisiensi waktu penanganan alamat secara signifikan. Rata-rata waktu interpretasi alamat pada proses manual berbasis pengetahuan kurir sebesar 3,33 menit dapat diturunkan menjadi 0,166 menit, sedangkan rata-rata waktu pencarian lokasi dari 9,44 menit menurun menjadi 0,125 menit. Dengan demikian, total waktu penanganan berkurang dari 12,78 menit menjadi 0,292 menit per alamat. Dari sisi spasial, deviasi jarak prediksi awal pada pendekatan manual sebesar 3.385 m dapat dipersempit menjadi 879 m setelah penerapan sistem hybrid, yang menunjukkan peningkatan konsistensi spasial hasil prediksi. Studi kasus Banjar Manyar dan Banjar Siyut menguatkan bahwa prediksi sistem koheren secara linguistik, hierarkis, dan spasial (mendekati titik pusat Banjar yang dituju). Hasil penelitian menegaskan kontribusi metodologis sistem prediksi alamat non-gazetteer berbasis pengetahuan yang relevan untuk wilayah dengan pengalamatan informal berbasis komunitas.
===============================================================================================================================
The efficacy of last-mile delivery is heavily reliant on the precise identification of residential addresses. In Bali, many addresses adhere to Banjar-based conventions that are not cataloged in formal geographic databases, thereby diminishing the effectiveness of gazetteer-based geocoding. This study introduces a knowledge-based system designed to predict the locations of Balinese residential addresses without the use of gazetteers by integrating multi-metric fuzzy matching with Hierarchical Reinforcement Learning based on Deep Q-Networks (HRL-DQN). The fuzzy matching component employs Levenshtein Distance, Partial Ratio, and Token Sort Ratio through a Hybrid Best-Score mechanism (Mix Score ranging from 0 to 100) to represent the linguistic similarity of unstructured address texts, while the HRL-DQN facilitates hierarchical predictions at the village (desa/kelurahan) and Banjar levels. An evaluation utilizing 17,456 historical shipment records from Pos Indonesia’s operational delivery data in Gianyar Regency indicates that the proposed workflow improves both temporal efficiency and spatial consistency in last-mile address handling. Under manual, courier knowledge–based processing, the average address interpretation time was 3.33 minutes and the average location search time was 9.44 minutes, yielding an average total handling time of 12.78 minutes per address. After applying the proposed hybrid Fuzzy Matching + HRL-DQN approach, the average interpretation time decreased to 0.166 minutes, the average location search/inference time decreased to 0.125 minutes, and the average total handling time was reduced to 0.292 minutes per address. In addition, the initial spatial deviation observed in the manual approach (3,385 m) was reduced to 879 m after applying the hybrid workflow, reflecting a more compact and geographically consistent set of predicted locations. Case studies conducted in Banjar Manyar and Banjar Siyut further validated that the system’s predictions were linguistically, hierarchically, and spatially coherent, converging toward the intended Banjar centroids. These findings substantiate the methodological contribution of a non-gazetteer, knowledge-based address prediction system that is particularly well suited for regions characterized by community-based and informal addressing practices.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | sistem berbasis pengetahuan; prediksi alamat; non-gazetteer; Banjar Bali; fuzzy matching; HRL-DQN; last-mile delivery. knowledge-based system; address prediction; non-gazetteer; Balinese Banjar; fuzzy matching; HRL-DQN; last-mile delivery. |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.6 Management information systems |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 55003-(S3) PhD Thesis |
| Depositing User: | Muhammad Isa Ansori |
| Date Deposited: | 27 Jan 2026 03:18 |
| Last Modified: | 27 Jan 2026 03:18 |
| URI: | http://repository.its.ac.id/id/eprint/130699 |
Actions (login required)
 |
View Item |