Shuffle based Anomaly Detection in Multi-state System

摘要: The anomaly events are defined as the points that rare and diverse from other in feature space. Conventional detection methods usually find low-probability with a learned probability distribution model, or evaluate testing samples local density of samples. Multi-state system has multiple normal states, state changes at unpredictable caused by daily operation such feed, outlet, flow control, etc. For multistate system, collecting enough data contain all possible states challenging impossible to users. Furthermore, conventional sensitive context training datasets unpredicted phased datasets, just consider Motivated this problem, we transform model learning problem distinction learns familiarity each In order reduce effects changes, randomly shuffle dataset use sliding window one-class Support Vector Machine (SVM) method. Our contributions include: (1) reducing requirement prior knowledge; (2) handling (3) considering global proposed method is evaluated on synthetic real experiments results show our superior than state-of-the-art methods.