<p>Density Peak Clustering (DPC) assumes that cluster centers are typically surrounded by points with lower local density and are at a significant distance from other centers. However, for complex datasets, the fixed parameter settings and point assignment strategies in DPC often misclassify irrelevant points, resulting in suboptimal clustering. To address this issue, we propose an adaptive parameter tuning strategy that automatically adjusts parameters based on the dataset’s characteristics. Additionally, we introduce a novel density region division method to effectively isolate noise points and simplify the clustering process by identifying potential true cluster centers in low-density regions. Furthermore, we refine the clustering process by merging sub-clusters through the shortest path in the graph, enhancing the accuracy of the final clusters. Experimental results on six synthetic and eight real-world datasets demonstrate that our algorithm not only outperforms the original DPC algorithm but also exceeds the performance of four other state-of-the-art DPC variants.</p>