Abstract
A well-known and adaptable classifier is the k-Nearest Neighbor (kNN) that requires a training set of relatively small size in order to perform adequately. Training sets can be reduced in size by using conventional data reduction techniques. Unfortunately, these techniques are inappropriate in streaming environments or when executed in devices with limited resources. dRHC is a prototype generation algorithm that works in streaming environments by maintaining a condensed training set that can be updated by continuously arriving training data segments. Prototypes in dRHC carry an appropriate weight to indicate the number of instances of the same class that they represent. dRHC2 is an improvement over dRHC since it can handle fixed size condensing sets by removing the least important prototypes whenever the condensing set exceeds a predefined size. In this paper, we exploit the idea that dRHC or dRHC2 prototypes could be merged whenever they are close enough and represent the same class. Hence, we propose two new prototype merging algorithms. The first algorithm performs a single pass over a newly updated condensing set and merges all prototype pairs of the same class under the condition that each prototype is the nearest neighbor of the other. The second algorithm performs repetitive merging passes until there are no prototypes to be merged. The proposed algorithms are tested against several datasets and the experimental results reveal that the single pass variation performs better for both dRHC and dRHC2 taking into account the trade-off between preprocessing cost, reduction rate and accuracy. In addition, the merging appears to be more appropriate for the static version of the algorithm (dRHC) since it offers higher data reduction without sacrificing accuracy.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Aggarwal, C.: Data Streams: Models and Algorithms. Advances in Database Systems Series, Springer, Boston (2007). https://doi.org/10.1007/978-0-387-47534-9
Cover, T., Hart, P.: Nearest neighbor pattern classification. IEEE Trans. Inf. Theory 13(1), 21–27 (2006). http://dx.doi.org/10.1109/TIT.1967.1053964
Ougiaroglou, S., Evangelidis, G.: RHC: a non-parametric cluster-based data reduction for efficient k-NN classification. Pattern Anal. Appl. 19(1), 93–109 (2014). http://dx.doi.org/10.1007/s10044-014-0393-7
Ougiaroglou, S., Arampatzis, G., Dervos, D.A., Evangelidis, G.: Generating fixed-size training sets for large and streaming datasets. In: Kirikova, M., Nørvåg, K., Papadopoulos, G.A. (eds.) ADBIS 2017. LNCS, vol. 10509, pp. 88–102. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-66917-5_7
Triguero, I., Derrac, J., Garcia, S., Herrera, F.: A taxonomy and experimental study on prototype generation for nearest neighbor classification. Trans. Sys. Man Cyber Part C 42(1), 86–100 (2012). http://dx.doi.org/10.1109/TSMCC.2010.2103939
Garcia, S., Derrac, J., Cano, J., Herrera, F.: Prototype selection for nearest neighbor classification: taxonomy and empirical study. IEEE Trans. Pattern Anal. Mach. Intell. 34(3), 417–435 (2012). http://dx.doi.org/10.1109/TPAMI.2011.142
Lozano, M.: Data Reduction Techniques in Classification Processes. Ph.D. Thesis, Universitat Jaume I (2007)
Ougiaroglou, S., Evangelidis, G.: Efficient dataset size reduction by finding homogeneous clusters. In: Proceedings of the Fifth Balkan Conference in Informatics, BCI 2012, ACM, New York, NY, USA, pp. 168–173 (2012). http://doi.acm.org/10.1145/2371316.2371349
Alcalá-Fdez, J., Fernández, A., Luengo, J., Derrac, J., García, S.: KEEL data-mining software tool: data set repository, integration of algorithms and experimental analysis framework. Multiple Valued Logic and Soft Comput. 17(2–3), 255–287 (2011)
Acknowledgments
This research is funded by the University of Macedonia Research Committee as part of the “Principal Research 2019” funding program.
We thank Prof. Yannis Manolopoulos for his excellent remarks during ADBIS 2017 that led to the ideas presented in this paper.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Ponos, P., Ougiaroglou, S., Evangelidis, G. (2019). Improving Data Reduction by Merging Prototypes. In: Welzer, T., Eder, J., Podgorelec, V., Kamišalić Latifić, A. (eds) Advances in Databases and Information Systems. ADBIS 2019. Lecture Notes in Computer Science(), vol 11695. Springer, Cham. https://doi.org/10.1007/978-3-030-28730-6_2
Download citation
DOI: https://doi.org/10.1007/978-3-030-28730-6_2
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-28729-0
Online ISBN: 978-3-030-28730-6
eBook Packages: Computer ScienceComputer Science (R0)