DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a clustering algorithm that has proven itself in data mining and machine learning. Among its advantages are its ability to identify clusters of arbitrary shapes, its robustness to noise, its adaptability to density variations, and its efficiency for large-dimensional datasets without worrying about specifying the number of clusters in advance. However, it can present difficulties concerning the sensitivity to two main parameters: ε (epsilon) and MinPts. Choosing these parameters can significantly impact the clustering results, and finding the optimal values can sometimes be challenging. This paper aims to improve the DBSCAN algorithm by proposing a new version, BN-DBSCAN (Best Noise-DBSCAN), to adjust the MinPts parameter. Rather than fixing this parameter to a given value, the proposed algorithm introduces the concept of “candidate noise points” (Best Noise) to help adaptively determine the value of MinPts. Using the idea of best noise points, BN-DBSCAN can extend existing clusters by including points initially considered noise, and new clusters can emerge from these best noise points. This is a novel approach to improve the DBSCAN algorithm by adaptively adjusting the MinPts parameter using the concept of best noise points, making the algorithm more robust and less dependent on prior domain knowledge or data configuration.