Hybrid Deep Learning Model for Time Series Anomaly Detection
Chan Sik Han, Keon Myung Lee · 2023
Multivariate time series anomaly detection is a fundamental challenge in real-world applications such as industry and business1. To address this issue, numerous models with diverse structures have been proposed. Each model leverages its unique structural characteristics to extract crucial features for time series anomaly detection. Our objective is to assess whether the performance of multivariate time series anomaly detection can be improved by employing a combination of models with different structures. In this paper, we propose a hybrid model for multivariate time series anomaly detection. The proposed hybrid model comprises two sub-models, each with a unique structure, and a simple layer. Each sub-model is designed to extract significant features from input time series. The simple layer combines the extracted features from both sub-models to generate the final output. Experimental results demonstrate that the proposed hybrid model outperforms single models.