Affiliations: [a] Veltech RR and SR Technical University, Avadi, Chennai, India | [b] Department of Computer Science and Engineering, Veltech RR and SR Technical University, Avadi, Chennai, India
Correspondence:
[*]
Corresponding author: Jagannath E. Nalavade, Veltech RR and SR Technical University, Avadi, Chennai, India. E-mail:jen20074u@gmail.com
Abstract: Due to the continuous growth of recent applications such
as, telecommunication, sensor data, financial applications, analyzing of
data streams, conceptually endless sequences of data records, frequently
arriving at high rates is important task in data mining. Among the various
tasks involved in data mining, the classification of data streams poses
various challenging issues as compared to popular algorithms of data
classification. Since the classification algorithm performs endlessly, it
must be able to adapt the classification model to handle the change of
concept or boundaries between classes. In order to handle these issues, we
have developed a new fuzzy system called, HRFuzzy for classification of
evolving data streams. Here, rough set theory and holoentropy function are
utilized to construct the dynamic classification model. In the fuzzy system,
the rules are generated using k-means clustering and membership functions
are dynamically updated using holoentropy function. The experimentation of
the proposed HRFuzzy is performed using two different databases such as,
skin segmentation dataset and localization data. The performance is compared
with the adaptive k-NN classifier in terms of accuracy and time. From the
outcome, we proved that the proposed HRFuzzy outperformed in both the
metrics by giving the maximum performance.
Keywords: Data stream, classification, fuzzy, rough set, holoentropy, concept change
