AnomalyBAGS: Bagging-based Ensemble Learning with Time-Series-to-Image Transformations for Robust Anomaly Detection

This work presents AnomalyBAGS, a novel method for improving the robustness of anomaly detection in time series data by using bagging-based ensemble learning combined with time-series-to-image transformations. With the increase of time series data in various domains, such as finance, healthcare, and network systems, the need for efficient anomaly detection is of major importance. Traditional methods often cannot handle the dynamics and nuances of time series data, so more advanced methods need to be explored. AnomalyBAGS harnesses the potential of ensemble learning by combining the strengths of multiple models to increase accuracy and performance. The method focuses on creating custom bagging ensembles by applying unique modifications to the training dataset of each member ensuring holistic consolidation of insights from multiple models. The results highlight the central role of selecting and combining the right data transformation techniques, with certain coding methods performing particularly well on specific datasets. Finally, this research not only advances the field of anomaly detection in time series data, but also provides a valuable reference tool for practitioners seeking to select the most appropriate data transformation methods for their specific datasets.