An Outlier Detection Algorithm based on Local Density and Natural Neighbors

Outlier detection is a hot research topic in data science, with the goal of identifying individuals in a dataset that are significantly different from other samples. This technique is widely used in fields such as image anomaly analysis and credit card fraud detection. Currently, there are many outlier detection algorithms available, but many of them struggle with the problem of difficult parameter selection and cannot simultaneously detect global and local outliers. To address this, we propose an outlier detection algorithm based on local density and natural neighbors (ODLDNN), which has the following advantages: (1) it inherits the non-parametric nature of natural neighbors and does not require the manual setting of parameter K; (2) it proposes a new method for calculating local density, which not only does not require parameters, but can also be used to detect both global and local outliers. Numerous experiments have shown that our algorithm can effectively detect both local and global outliers without the need for setting the neighborhood parameter K.