Aiming at the problems of the classical DBSCAN algorithm that it is difficult to determine the global optimal parameters and misjudge the outliers, the algorithm firstly starts from the perspective of selecting the optimal parameters, generates the Eps list and the MinPts list by the distribution characteristics of the data set, and then carries out the full combination operation of the parameters in the two lists, and then performs the clustering of the different combinations of the parameters in order, so as to search for the parameter corresponding to the point of the highest accuracy rate. Finally, from the perspective of outlier, the three-branch decision-making idea is combined with the outlier detection LOF algorithm. This algorithm is compared and analysed with a variety of clustering algorithms, and the results show that this algorithm can achieve the fully automated selection of the global optimal parameters as well as improve the accuracy of the clustering algorithm.

[2024-01-31] Accepted Paper /doc/2023.11.0569/v1