An efficient method for autoencoder based outlier detection

作者：

Highlights：

• Two techniques are introduced to detecting outliers by exploiting Autoencoder.

• Density and distance features of each point are used to detect probable outliers.

• Having removed probable outliers from the dataset, an Autoencoder is applied.

• Finally, the obtained Autoencoder is used for finding outlier from whole dataset.

• Experimental results demonstrate superiority of proposed methods on real datasets.

摘要

•Two techniques are introduced to detecting outliers by exploiting Autoencoder.•Density and distance features of each point are used to detect probable outliers.•Having removed probable outliers from the dataset, an Autoencoder is applied.•Finally, the obtained Autoencoder is used for finding outlier from whole dataset.•Experimental results demonstrate superiority of proposed methods on real datasets.

论文关键词：Outlier detection,Self organizing map (SOM),Autoencoder,Density peaks clustering (DPC),Randomized neural network for outlier detection (RandNet),One class support vector machine (OCSVM),Boosting-based autoencoder ensemble (BAE)

论文评审过程：Received 10 February 2022, Revised 23 September 2022, Accepted 23 September 2022, Available online 29 September 2022, Version of Record 10 October 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118904