Abstract:This paper introduces a novel anomaly detection approach tailored for time series data with exclusive reliance on normal events during training. Our key innovation lies in the application of kernel-density estimation (KDE) to scrutinize reconstruction errors, providing an empirically derived probability distribution for normal events post-reconstruction. This non-parametric density estimation technique offers a nuanced understanding of anomaly detection, differentiating it from prevalent threshold-based mechanisms in existing methodologies. In post-training, events are encoded, decoded, and evaluated against the estimated density, providing a comprehensive notion of normality. In addition, we propose a data augmentation strategy involving variational autoencoder-generated events and a smoothing step for enhanced model robustness. The significance of our autoencoder-based approach is evident in its capacity to learn normal representation without prior anomaly knowledge. Through the KDE step on reconstruction errors, our method addresses the versatility of anomalies, departing from assumptions tied to larger reconstruction errors for anomalous events. Our proposed likelihood measure then distinguishes normal from anomalous events, providing a concise yet comprehensive anomaly detection solution. The extensive experimental results support the feasibility of our proposed method, yielding significantly improved classification performance by nearly 10% on the UCR benchmark data.

Anomaly Detection With Kernel Preserving Embedding

Anomaly Detection with Representative Neighbors.

Robust and Efficient Anomaly Detection Using Heterogeneous Representations

An Embedding Approach to Anomaly Detection

An Anomaly Detection Approach Based On Symbolic Similarity

Anomaly Detection and Localization in Crowded Scenes by Motion-field Shape Description and Similarity-based Statistical Learning

Anomaly Detection Using Local Kernel Density Estimation and Context-Based Regression.

An Anomaly Detection Method for Hyperspectral Imagery in Kernel Feature Space Based on Robust Analysis

Similarity Learning Via Kernel Preserving Embedding

Kernel-Based Nonlinear Anomaly Detection Via Union Dictionary for Hyperspectral Images

Detecting Anomalies in Time Series Using Kernel Density Approaches

Anomaly Detection In Hyperspectral Imagery Based On Low-Rank Representation Incorporating A Spatial Constraint

Anomaly Detection Method Based on Temporal Spatial Information Enhancement

Research of Anomaly detection based on Dynamic Anomaly Detection Enhancement Framework

Adaptive Kernel Density-Based Anomaly Detection for Nonlinear Systems

k-NNN: Nearest Neighbors of Neighbors for Anomaly Detection

Isolation Distributional Kernel: A New Tool for Kernel Based Anomaly Detection

Anomaly detection using large-scale multimode industrial data: An integration method of nonstationary kernel and autoencoder

Anomaly detection using sparse reconstruction in crowded scenes

Video Anomaly Detection Based on Deep Learning

Anomaly Detection By Analyzing The Pedestrian Behavior And The Dynamic Changes Of Behavior