Search

In this study, we present a clustering cost function that uses K-nearest neighbour (KNN) and particle swarm optimisation (PSO) to detect outliers. Here, the Dunn index was used to estimate the ideal number of clusters. The obtained number of clusters was subsequently utilised to cluster data using the PSO algorithm and our proposed cost function. The KNN method was used to determine the top n outliers in the datasets, and data pruning was performed to eliminate any potential inliers. Outlier detection average precision was measured using both the K-means and our proposed algorithm. It was discovered that our proposed technique consistently detects outliers across different types of datasets containing both extremely and marginally distinct outliers from inliers data points. We believe that our proposed algorithm could be a promising technique for the outlier detection problem, with applications in network intrusion, medical diagnosis, banking, cyber-security, optical communication, etc. [ABSTRACT FROM AUTHOR]