Authors:
Camelia Lemnaru
;
Andreea Tudose-Vintila
;
Andrei Coclici
and
Rodica Potolea
Affiliation:
Technical University of Cluj-Napoca, Romania
Keyword(s):
Imbalanced Multi-class Classification, Multiple Classifier, Prediction Combination, Anomaly Detection, KDD’99 Data.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Clustering and Classification Methods
;
Data Reduction and Quality Assessment
;
Knowledge Discovery and Information Retrieval
;
Knowledge-Based Systems
;
Mining High-Dimensional Data
;
Pre-Processing and Post-Processing for Data Mining
;
Symbolic Systems
Abstract:
Imbalanced classification problems represent a current challenge for the application of data mining techniques to real-world problems, since learning algorithms are biased towards favoring the majority class(es). The present paper proposes a compound classification architecture for dealing with imbalanced multi-class problems. It comprises of a two-level classification system: a multiple classification model on the first level, which combines the predictions of several binary classifiers, and a supplementary classification model, specialized on identifying “difficult” cases, which is currently under development. Particular attention is allocated to the pre-processing step, with specific data manipulation operations included. Also, a new prediction combination strategy is proposed, which applies a hierarchical decision process in generating the output prediction. We have performed evaluations using an instantiation of the proposed model applied to the field of network intrusion detect
ion. The evaluations performed on a dataset derived from the KDD99 data have indicated that our method yields a superior performance for the minority classes to other similar systems from literature, without degrading the overall performance.
(More)