Latent Vector Expansion using Autoencoder for Anomaly Detection

TL;DR

This paper introduces a latent vector expansion autoencoder that enhances anomaly detection performance in imbalanced datasets by augmenting features and improving feature distinction.

Contribution

The paper proposes a novel latent vector expansion autoencoder model that improves anomaly detection accuracy in imbalanced datasets by augmenting and distinguishing features.

Findings

Improved anomaly detection performance over basic autoencoders.

Effective feature augmentation for imbalanced datasets.

Enhanced classification accuracy in real-world anomaly detection scenarios.

Abstract

Deep learning methods can classify various unstructured data such as images, language, and voice as input data. As the task of classifying anomalies becomes more important in the real world, various methods exist for classifying using deep learning with data collected in the real world. As the task of classifying anomalies becomes more important in the real world, there are various methods for classifying using deep learning with data collected in the real world. Among the various methods, the representative approach is a method of extracting and learning the main features based on a transition model from pre-trained models, and a method of learning an autoencoderbased structure only with normal data and classifying it as abnormal through a threshold value. However, if the dataset is imbalanced, even the state-of-the-arts models do not achieve good performance. This can be addressed by augmenting normal and abnormal features in imbalanced data as features with strong distinction. We use the features of the autoencoder to train latent vectors from low to high dimensionality. We train normal and abnormal data as a feature that has a strong distinction among the features of imbalanced data. We propose a latent vector expansion autoencoder model that improves classification performance at imbalanced data. The proposed method shows performance improvement compared to the basic autoencoder using imbalanced anomaly dataset.