An intelligent prediction system for time series data using periodic pattern mining in temporal databases

  • Authors:
  • S. Sridevi;S. Rajaram;C. Swadhikar

  • Affiliations:
  • Thiagarajar College of Engineering, Madurai;Thiagarajar College of Engineering, Madurai;Thiagarajar College of Engineering, Madurai

  • Venue:
  • Proceedings of the First International Conference on Intelligent Interactive Technologies and Multimedia
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Data mining is concerned with analyzing large volumes of unstructured data to discover interesting regularities or relationships which in turn lead to better understanding of the underlying processes. Existing algorithms like association rule mining, incremental mining and frequent pattern mining can be used to find out valid periodic pattern and it can't be used to find out peculiar data. In this paper, two algorithms namely Peculiarity factor algorithm and Chi-Square test algorithm are used to find out peculiar data from a temporal database which is presented in vertical format. If peculiar data are found in two different relations there is need to use a value in a key as the relevance factor in order to find out the relevance between those relations. Thus a new dataset is formed from an existing dataset after the removal of peculiar data. From a new dataset Periodic Patterns were found by applying four phase algorithms namely singular periodic pattern mining, multi-event periodic pattern mining, complex periodic pattern mining and asynchronous sequence mining. Our proposed work focuses on prediction of time series data. This can be done with the help of correlation estimation. After determining strong and weak attributes using correlation estimation only strong attributes are considered to find out how each attribute is correlated with other attributes. Based on the correlation we predicted the required attribute values under given test conditions. Based on the prediction output precision and recall are calculated and hence accuracy is measured. Experimental results on real-life datasets demonstrate that the proposed algorithm is effective and efficient to predict the time series data.