Analysis of PT PLN (Persero)'s New Installation Waiting List Using the K-Means Clustering Algorithm

Figure Ilustration AI

FORMOSA NEWS - Medan - PT PLN (Persero) Leverages K-Means Clustering to Optimize Indonesian Power Grid Allocation and Streamline Public Services. Managing a national electricity grid across thousands of islands is an immensely complex logistical challenge. To maintain public trust and comply with strict national service regulations, energy providers must process new electricity connection requests rapidly. A groundbreaking joint study conducted by researchers Ernawati and Dewi Agushinta R. from Universitas Gunadarma has successfully applied advanced data science to solve this issue. By using the K-Means clustering algorithm, the researchers mapped out data from the final three months of 2024 to identify which regional branches of PT PLN (Persero) face the heaviest operational bottleneck and backlog. The findings provide a robust, data-driven framework that allows management to strategically allocate human and technical resources exactly where they are needed most.

Simplifying Grid Operational Data via RapidMiner
To cut through this analytical clutter, the Universitas Gunadarma research team adopted a quantitative data-mining approach utilizing K-Means clustering. Conducted between December 2024 and January 2025, the study extracted and transformed secondary data directly from the internal databases of PT PLN (Persero). The raw data was carefully filtered to include only valid, finalized applications for new power installations. By feeding this structured information into the RapidMiner software platform, the algorithm partitioned the 22 regional units into three distinct clusters based on their total volume of backlog: high volume (C1), medium volume (C2), and low volume (C3). To ensure the resulting clusters were statistically sound and legally accountable, the researchers ran the model through the Davies-Bouldin Index (DBI). The algorithm achieved an exceptionally stable DBI score of 0.486. In data science, a lower DBI score close to zero confirms that the boundaries between data groups are sharp, distinct, and cohesive, proving that the study's findings are highly accurate and completely free from random statistical noise.

Key Findings: Java Island Segments Suffer Extreme Connection Bottlenecks
The K-Means clustering algorithm revealed a severe, disproportionate imbalance in the geographical distribution of Indonesia's utility workload. The 22 distribution units were divided as follows:

• Cluster 1 (High Backlog Volume - 6 Units): This group represents the highest service pressure nationwide. The East Java distribution unit (UID East Java) experienced the single worst bottleneck in the country, accumulating a massive 69,892 pending requests. It was closely followed by UID Central Java & Yogyakarta with 57,851 requests; UIW South Sulawesi & West Sulawesi with 56,870 requests; UIW S2JB (covering South Sumatra, Jambi, and Bengkulu) with 47,516 requests; UID West Java with 42,957 requests; and UIW South and Central Kalimantan with 41,755 requests.
  
• Cluster 2 (Medium Backlog Volume - 7 Units): These regions maintain a stable balance between incoming consumer demand and operational capacity but remain vulnerable to future economic shocks. This cluster contains UIW Central Sulawesi (28,146), UID Bali (27,127), UIW East Nusa Tenggara (25,869), UIW North Sumatra (23,677), UIW West Kalimantan (20,728), UID Lampung (19,912), and UIW Papua & West Papua (19,166).
  
• Cluster 3 (Low Backlog Volume - 9 Units): Representing areas with minimal service strain where current capacity easily meets local demand, this cluster features UIW Riau and Riau Islands (17,053), UID Greater Jakarta (16,848), UIW West Nusa Tenggara (14,192), UIW Maluku & North Maluku (11,549), UIW East Kalimantan (11,188), UID Banten (10,863), UIW Aceh (7,382), UIW West Sumatra (7,202), and UIW Bangka Belitung (2,580).
  

Beyond the regional groupings, the study evaluated PT PLN (Persero)'s performance against its own Service Quality Targets (TMP). The data indicated that 55 percent of all new installation requests were processed within the ideal 5-day standard without requiring any grid expansion. However, 45 percent of all national applicants faced significant delays, failing to get connected within the mandatory 5-day timeframe due to severe underlying infrastructure constraints.

Real-World Impact and Grid Policy Implications
The practical implications of this Universitas Gunadarma study are highly valuable for policymakers, public utility managers, and businesses waiting on grid connectivity. By isolating the six critical units in Cluster 1, the research eliminates guesswork for PT PLN (Persero) executives. Instead of scattering resources uniformly across the country, management can deploy emergency technical personnel, specialized physical equipment, and streamlined administrative support directly to high-pressure zones like East Java and Central Java to avoid heavy regulatory penalties. The researchers noted that the 45 percent of delayed connections are not caused by simple employee negligence but are tied to structural problems. In high-demand zones, rapid urbanization and commercial expansion require complex field work, including physical grid expansions, navigating difficult local geography, and securing additional regulatory permits.

Author Profiles
Ernawati holds an advanced academic degree in Computer Science from Universitas Gunadarma. She is a specialized data scientist focusing on computational modeling, algorithmic data mining, and the digital optimization of public infrastructure workflows.
Dewi Agushinta R. is a senior professor and prominent computer science researcher affiliated with Universitas Gunadarma. Her deep academic expertise spans artificial intelligence design, non-hierarchical clustering algorithms, and the integration of data science frameworks within corporate and industrial management.

Source
Ernawati, Dewi Agushinta R 2026, Analysis of PT PLN (Persero)'s New Installation Waiting List Using the K-Means Clustering Algorithm. Formosa Journal of Computer and Information Science (FJCIS) 2026. Vol. 5, No. 1, Halaman 63-82
DOI: https://doi.org/10.55927/fjcis.v5i1.16429
URL: https://journal.formosapublisher.org/index.php/fjcis