Kepadatan Dealer, Spatial Misallocation, dan Potensi Pasar: Studi Analitika Bisnis Industri Otomotif Indonesia

Marfelly, Mario (2026) Kepadatan Dealer, Spatial Misallocation, dan Potensi Pasar: Studi Analitika Bisnis Industri Otomotif Indonesia. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032241187-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Distribusi jaringan dealer otomotif di Indonesia belum sepenuhnya selaras dengan kapasitas permintaan, ekonomi, aksesibilitas, dan keterkaitan spasial wilayah. Kondisi tersebut menimbulkan spatial misallocation. Penelitian ini bertujuan untuk menganalisis pola distribusi dealer, memprediksi kesenjangan jaringan dan kategori pasar, membandingkan kinerja pendekatan pemodelan, serta merumuskan prioritas ekspansi jaringan dealer otomotif di Indonesia. Penelitian menggunakan data panel 514 kabupaten/kota selama periode 2018–2024 yang mencakup variabel ekonomi, demografi, pasar, aksesibilitas, dan karakteristik wilayah. Metode yang digunakan meliputi Exploratory Spatial Data Analysis (ESDA), Spatial Econometrics melalui Spatial Lag Model (SLM), Ensemble Machine Learning menggunakan Random Forest dan XGBoost, serta Graph Neural Network melalui Graph Convolutional Network (GCN) dan Graph Attention Network (GAT). ESDA dan Spatial Econometrics digunakan untuk menganalisis kondisi cross-section tahun 2024, sedangkan Machine Learning dan GNN dievaluasi melalui validasi temporal dan desain prediksi tahun berikutnya. Hasil menunjukkan bahwa dependensi spasial pada distribusi dealer relatif lemah dan sensitif terhadap definisi ketetanggaan. Pada variabel Dealer Gap, SLM menghasilkan koefisien spatial lag moderat sebesar −0.283 (p<0,001). Hasil tersebut mengindikasikan adanya pola spatial antar wilayah. Berdasarkan validasi, XGBoost Versi 1 terpilih sebagai model regresi utama dan menghasilkan R^2sebesar 0.81. Random Forest Versi 1 terpilih sebagai model klasifikasi utama dengan recall Expansion Potential sebesar 0.93. GCN Queen Contiguity menghasilkan performa lebih baik dibandingkan GAT, tetapi keduanya masih berada di bawah model tree-based dan digunakan sebagai validasi struktural berbasis graf. Hasil seluruh metode diintegrasikan ke dalam Dealer Expansion Priority Matrix. Threshold berbasis kuantil Q95 menghasilkan 26 wilayah pada Population Baseline dengan nilai 0.7551 dan 26 wilayah pada spesifikasi SLM Gap Score dengan nilai 0.7721. Sebanyak 25 wilayah muncul secara konsisten pada kedua spesifikasi dan ditetapkan sebagai prioritas final. Hasil tersebut menunjukkan bahwa ekspansi diposisikan sebagai proses bertingkat dan disesuaikan dengan kelayakan bisnis.
=================================================================================================================================
The distribution of automotive dealer networks in Indonesia is not yet fully aligned with regional demand capacity, economic conditions, accessibility, and spatial interrelationships. This condition gives rise to spatial misallocation. This study aims to analyze dealer distribution patterns, predict network gaps and market categories, compare the performance of different modeling approaches, and formulate priorities for automotive dealer network expansion in Indonesia. The study uses panel data covering 514 districts and municipalities during the 2018–2024 period, including economic, demographic, market, accessibility, and regional characteristics. The methods employed include Exploratory Spatial Data Analysis (ESDA), spatial econometrics using the Spatial Lag Model (SLM), ensemble machine learning using Random Forest and XGBoost, and Graph Neural Networks through the Graph Convolutional Network (GCN) and Graph Attention Network (GAT). ESDA and spatial econometrics are used to analyse the 2024 cross-sectional condition, while the machine learning and GNN models are evaluated through temporal validation and a next-year prediction design. The results indicate that spatial dependence in dealer distribution is relatively weak and sensitive to the definition of neighbourhood relationships. For the Dealer Gap variable, the SLM produces a moderate spatial lag coefficient of −0.283 (p < 0.001). This finding indicates the presence of spatial patterns across regions. Based on the validation results, XGBoost Version 1 is selected as the primary regression model and achieves an R^2of 0.81. Random Forest Version 1 is selected as the primary classification model, with an Expansion Potential recall of 0.93. The GCN model using Queen Contiguity performs better than GAT; however, both models remain below the ML and are therefore used as graph-based structural validation. The outputs of all methods are integrated into the Dealer Expansion Priority Matrix. The Q95 quantile-based threshold identifies 26 regions under the Population Baseline specification at a threshold value of 0.7551 and 26 regions under the SLM Gap Score specification at a threshold value of 0.7721. A total of 25 regions consistently appear in both specifications and are designated as the final priority areas. These findings indicate that expansion should be positioned as a tiered process and adjusted according to business feasibility.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kata Kunci: analitika bisnis, spatial misallocation, dealer otomotif, machine learning, graph neural network, spatial econometrics Keywords: automotive dealers, business analytics, graph neural network, machine learning, spatial econometrics
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	61101-Magister Management Technology
Depositing User:	Mario Marfelly
Date Deposited:	30 Jul 2026 03:51
Last Modified:	30 Jul 2026 03:51
URI:	http://repository.its.ac.id/id/eprint/139596

Actions (login required)

View Item