# Application of PSO-integrated K-means algorithm in resident digital portrait classification

## TL;DR

This paper introduces a new PSO-K-means algorithm to improve the classification of resident digital portraits, enhancing digital governance at the community level.

## Contribution

The novel PSO-KM algorithm combines Particle Swarm Optimization with K-means for better clustering of high-dimensional resident data.

## Key findings

- PSO-KM achieved a silhouette score of 0.752 ± 0.021 and inter-cluster distance of 1.493 ± 0.036, outperforming traditional methods.
- Behavioral data showed the highest classification performance with a silhouette value of 0.184.
- Dominant features vary by income bracket, with behavioral metrics most influential for middle-income groups.

## Abstract

As digital governance progresses rapidly, constructing digital portraits of residents has become instrumental in enhancing local-level administrative capabilities. Nonetheless, traditional K-means clustering algorithms struggle with the classification of high-dimensional and complex data, thereby limiting their effectiveness. To address this issue, this paper proposes a novel hybrid algorithm—PSO-KM—that integrates Particle Swarm Optimization with K-means to improve both accuracy and computational efficiency in clustering resident profile data. Drawing on comprehensive resident information collected in 2023 from a community management system, the method leverages PSO’s global optimization abilities alongside K-means’ iterative refinement to dynamically update cluster centroids. Performance evaluation shows a significant uplift in clustering metrics, with a silhouette score of 0.752 ± 0.021 and inter-cluster distance of 1.493 ± 0.036. Comparative analysis against conventional and advanced methods (e.g., GA-K-means, DBSCAN) reveals that PSO-KM delivers superior outcomes. Among different feature categories, behavioral data yield the best classification performance, with a silhouette value of 0.184, highlighting the discriminatory power of dynamic behavioral traits. Furthermore, segmentation results disclose varying dominant features across income brackets: demographic factors are primary for low-income groups, behavioral metrics dominate middle-income segments, while social network indicators are key for high-income populations. These insights confirm PSO-KM’s potential in refining digital profiling processes and fostering the advancement of grassroots digital governance practices.