Abnormality detection method based on ranking aggregation and stacking

The invention relates to the technical field of anomaly detection, and particularly discloses a ranking aggregation and stacking-based anomaly detection method, which comprises the following steps of: enriching a feature space and injecting diversity into a model by using various unsupervised anomaly detection algorithms, and sorting and aggregating abnormal point scores to obtain an abnormal value matrix O (Xtrain), so that variance can be reduced. And then, integrating an aggregated abnormal value matrix O (Xtrain) by adopting a ranking aggregation abnormal score method to obtain an abnormal value division matrix as a pseudo tag of a training set Xtrain, thereby converting an unsupervised problem into a supervised learning task. And finally, a stacking-based dynamic classifier selection integration model is constructed, training and testing are carried out, and a relatively low variance can be obtained under the condition that too much deviation is not introduced. In a word, the method reduces the deviation