Intricate Anomaly Detection in Time Series by Caculating Shift of Non-Gaussianity

Time series serves as a fundamental concept in statistics, data analysis, and machine learning, permeating diverse domains like energy, neurology, Internet of Things (IoT), finance, economics, and environmental science. Understanding temporal patterns, especially in anomaly detection (AD) within time series, proves valuable yet challenging. There are three main approaches for AD: the statistical approach, cost-effective for revealing simple outliers; decomposition-based and learning-based approaches, expensive but proficient at identifying intricate patterns. This paper introduces an economical statistical method for detecting intricate anomalies by introducing a novel metric measuring the shift of non-Gaussianity. Unlike computationally intensive learning-based AD, the proposed algorithm efficiently identifies intricate anomalies by calculating the sum of absolute values of the derivatives of kurtosis and skewness. Its efficiency in both time and resource utilization makes it suitable for real-time and resource-constrained scenarios.