Abnormality Forecasting: Time Series Anomaly Prediction via Future Context Modeling

TL;DR

This work introduces a novel principled approach to future context modeling, FCM, aiming at providing early warnings for abnormal events before their occurrence, and demonstrates that FCM gains good recall rate on multiple datasets and significantly outperforms all baselines in F1 score.

Abstract

Identifying anomalies from time series data plays an important role in various fields such as infrastructure security, intelligent operation and maintenance, and space exploration. Current research focuses on detecting the anomalies after they occur, which can lead to significant financial/reputation loss or infrastructure damage. In this work we instead study a more practical yet very challenging problem, time series anomaly prediction, aiming at providing early warnings for abnormal events before their occurrence. To tackle this problem, we introduce a novel principled approach, namely future context modeling (FCM). Its key insight is that the future abnormal events in a target window can be accurately predicted if their preceding observation window exhibits any subtle difference to normal data. To effectively capture such differences, FCM first leverages long-term forecasting models to generate a discriminative future context based on the observation data, aiming to amplify those subtle but unusual difference. It then models a normality correlation of the observation data with the forecasting future context to complement the normality modeling of the observation data in foreseeing possible abnormality in the target window. A joint variate-time attention learning is also introduced in FCM to leverage both temporal signals and features of the time series data for more discriminative normality modeling in the aforementioned two views. Comprehensive experiments on five datasets demonstrate that FCM gains good recall rate (70\%+) on multiple datasets and significantly outperforms all baselines in F1 score. Code is available at https://github.com/mala-lab/FCM.

Figures (5)

• Figure 1: Illustration of the key insight of FCM. For effective anomaly prediction, FCM assumes that there are subtle signs (in the bottom time series) at the observation window for future abnormal events in a target window. These signs are typically too subtle to be detected by TSAD models. FCM aims to leverage long-term time series forecasting models to amplify these signs and associate them as a future context with the observation temporal signals for accurate anomaly prediction.
• Figure 2: The studied setting. Given an observational time window from $t_i$ to $t_{i+1}$, we aim to predict whether there would be abnormal events in the target window from $t_{i+1}$ to $t_{i+2}$. The two windows are slid with a fixed step size (see Sec. \ref{['subsec:setting']}).
• Figure 3: Overview of FCM. The future context-aware anomaly prediction in FCM leverages the output of the forecasting task to enhance the representation of the current window, generating a new discriminative representation of the observation data for joint normality modeling. The var-time attention module learns the relationships between temporal signals and feature values of time series data to enable more discriminative normality learning.
• Figure 4: Visualization of ground-truth anomalies and anomaly scores for different types of anomalies.
• Figure 5: Sensitivity analysis of main hyper-parameters in FCM.