Discovering an Evolutionary Classifier over a High-speed Nonstatic Stream

With the emergence of large-volume and high-speed streaming data, mining data streams has become a focus of increasing interest. The major new challenges in streaming data mining are as follows: since streams may flow in and out indefinitely and at fast speed, it is usually expected that a stream-mining process can only scan a data stream once; and since the characteristics of the data may evolve over time, it is desirable to incorporate the evolving features of data streams. This paper investigates the issues of developing a high-speed classification method for streaming data with concept drifts. Among several popular classification techniques, the naive Bayesian classifier is chosen due to its low construction cost, ease of incremental maintenance, and high accuracy. An efficient algorithm, called EvoClass (Evolutionary Classifier), is devised. EvoClass builds an incremental, evolutionary Bayesian classifier on streaming data. A train-and-test method is employed to discover the changes in the characteristics of the data and the need for construction of a new classifier. In addition, divergence is utilized to quantify the changes in the classifier and inform the user what aspects of the data characteristics have evolved. Finally, an intensive empirical study has been performed that demonstrates the effectiveness and efficiency of the EvoClass method.

[1]  Ian Witten,et al.  Data Mining , 2000 .

[2]  Sudipto Guha,et al.  Streaming-data algorithms for high-quality clustering , 2002, Proceedings 18th International Conference on Data Engineering.

[3]  JOHANNES GEHRKE,et al.  RainForest—A Framework for Fast Decision Tree Construction of Large Datasets , 1998, Data Mining and Knowledge Discovery.

[4]  David G. Stork,et al.  Pattern Classification , 1973 .

[5]  Philip M. Long.  The Complexity of Learning According to Two Models of a Drifting Environment , 2004, Machine Learning.

[6]  Rajeev Rastogi,et al.  Processing complex aggregate queries over data streams , 2002, SIGMOD '02.

[7]  Geoff Hulten,et al.  Mining high-speed data streams , 2000, KDD '00.

[8]  Johannes Gehrke,et al.  Querying and mining data streams: you only get one look a tutorial , 2002, SIGMOD '02.

[9]  Ian H. Witten,et al.  Data mining: practical machine learning tools and techniques with Java implementations , 2002, SGMD.

[10]  Jennifer Widom,et al.  Models and issues in data stream systems , 2002, PODS.

[11]  Yoav Freund,et al.  A decision-theoretic generalization of on-line learning and an application to boosting , 1997, EuroCOLT.

[12]  Geoff Hulten,et al.  Mining time-changing data streams , 2001, KDD '01.

[13]  Yixin Chen,et al.  Multi-Dimensional Regression Analysis of Time-Series Data Streams , 2002, VLDB.

[14]  Johannes Gehrke,et al.  BOAT—optimistic decision tree construction , 1999, SIGMOD '99.

[15]  Jiawei Han,et al.  Data Mining: Concepts and Techniques , 2000 .

[16]  Divesh Srivastava,et al.  On computing correlated aggregates over continual data streams , 2001, SIGMOD '01.

[17]  Sudipto Guha,et al.  Clustering data streams , 2000, Proceedings 41st Annual Symposium on Foundations of Computer Science.

[18]  Huan Liu,et al.  Discretization: An Enabling Technique , 2002, Data Mining and Knowledge Discovery.

[19]  Rajeev Motwani,et al.  Approximate Frequency Counts over Data Streams , 2012, VLDB.

[20]  Johannes Gehrke,et al.  A framework for measuring changes in data characteristics , 1999, PODS '99.

[21]  Jianhua Lin,et al.  Divergence measures based on the Shannon entropy , 1991, IEEE Trans. Inf. Theory.

[22]  Pedro M. Domingos,et al.  On the Optimality of the Simple Bayesian Classifier under Zero-One Loss , 1997, Machine Learning.