DeepSwarm-DBSCAN: Deep Learning and Swarm Intelligence for Automated Density-Based Clustering

Density-Based Spatial Clustering of Applications with Noise is widely used for unsupervised clustering due to its ability to identify arbitrarily shaped clusters and handle noise. However, its effectiveness strongly depends on appropriate hyperparameter selection, which becomes challenging and computationally expensive for high-dimensional data. This study aims to address this limitation by proposing an automated and scalable optimization framework. The proposed DeepSwarm-DBSCAN framework integrates deep neural networks with an enhanced Particle Swarm Optimization (PSO) algorithm to automatically determine optimal DBSCAN hyper- parameters. Convolutional Neural Networks (CNN) and Autoencoders (AE) learn compact and meaningful feature representations, projecting high-dimensional data into lower-dimensional latent spaces. An adaptive PSO strategy then searches this reduced space to optimize clustering validity metrics efficiently. Experimental evaluations on MNIST and CIFAR-10 datasets demonstrate that DeepSwarm-DBSCAN outperforms conventional DBSCAN, achieving clustering accuracies of 92%, 91%, and 94%—with a silhouette coefficient of 0.7 on MNIST—indicating strong cluster coherence. Additionally, the proposed approach reduces hyperparameter optimization runtime by approximately 30% compared to traditional grid search methods. Conclusion: The results confirm that combining deep representation learning with swarm-based optimization provides an effective solution for DBSCAN hyperparameter tuning. DeepSwarm- DBSCAN offers a robust, scalable, and efficient clustering framework suitable for real- world high-dimensional data analysis across diverse application domains. The results confirm that combining deep representation learning with swarm-based optimization provides an effective solution for DBSCAN hyperparameter tuning. DeepSwarm-DBSCAN offers a robust, scalable, and efficient clustering framework suitable for real-world high- dimensional data analysis across diverse application domains.