Research On Classification Method And Application Of Concept Drift Time Series Data

Time series data plays an important role in the research of data mining.The feature distribution of time series data is time-sensitive.When the distribution of the input time-series data stream changes greatly,concept drift occurs,which is one of the key factors affecting the performance of the classifier.Conceptual drift always exists in the production and application of time series data,the classification research of concept drift time series data has become one of the hotspots in the field of time series data mining.This paper systematically summarizes the research status of concept drift time series data at home and abroad,elaborates the time series data description method and mainstream data classification methods in detail,and conducts targeted research on the concept drift problem in time series data classification.All research work mainly includes:1.Research on the classification method of time series data based on the concept drift of the improved Hoeffding inequality.Aiming at the classification problem of time series data with concept drift,this paper proposes a concept drift detection method based on the interquartile range overlapping sliding window.This method uses the samples in the interquartile range window and the improved Hoeffding inequality to detect the concept drift.In order to better avoid the effect of noise on the performance of the classifier,the algorithm introduces a dynamic coefficient based on the classification accuracy of the current sample in the Hoeffding inequality to adapt to the change of different feature distributions of the data stream.2.Aiming at the problem of unbalanced concept drift time series data classification,a dynamic ensemble classifier model with adaptive adjustment of the basic classifier is proposed.This model proposes a new weight calculation method under the basic ensemble classifier,and So choose to join the basic classifier of the ensemble model.It also aims to reduce the spatial complexity of the ensemble classifier,the paper uses the classifier weight threshold elimination mechanism to delete the basic classifiers in the ensemble classifier that are not suitable for the current data stream feature distribution in time,so as to achieve the number of basic classifiers in the ensemble classification model.3.Apply the research done to spam applications.In order to test the effectiveness of the research results of this paper in spam classification and consider the imbalanced characteristics of spam classification data,this paper applies the proposed ensemble classifier model to spam classification.Experiments show that the ensemble classifier proposed in this paper can Improve the accuracy of spam classification more effectively.