A Gnana Baskaran


Clustering high dimensional spatial data for Forest fire risk analysis has been major issue due to sparsity of data points.. Most of the clustering algorithm becomes inefficient if the required distance similarity measure is computed for low dimensional spatial space of high dimensional data with sparsity of data point along different dimensions and also considering the obstacles.. Objective of this study were to contribute the complexity of projecting clusters for traffic risk analysis, (i) lack of support for reducing the number of dimensions on spatial space to reduce the searching time (ii) the lack of support for obstacles in the spatial data space. (iii) Compare computation time of HARP, Proclus, Doc, FastDoc, SSPC algorithms. Approach: During the first phase the satellite captured still images for different dimensions such as time and location of the forest fire network are enhanced and this images are given as input to red color image separation, During this phase the input images groped based on red color using K-Means algorithm and during the second phase the red color images are converted to gray scale images . The third phase mainly focuses on spatial attribute relevance analysis for detecting dense and sparse forest fire regions after detecting dense and sparse fire regions the algorithm employees pruning technique to reduce the search space by taking only dense fire regions and eliminating sparse fire regions and during fourth phase K-mediods algorithm is employed to project the clusters on different spatial dimensions and also it solves the problem of obstacles Results: First we showed that various projecting clustering algorithm on spatial space becomes inefficient if the number of dimensions increases .The new scheme proposed reduces the spatial dimension space so that it reduces the computation time and also it solves the problem of obstacles using K-mediods algorithim and finally the result is compared with HARP ,Proclus,Doc,FastDoc,SSPC The algorithms produces acceptable results when the average cluster dimensionality is greater than 10%. Conclusion: Hence the findings suggested the overhead reasonably minimized and using simulations, we investigated the efficiency of our schemes in supporting high dimensional spatial clustering for forest fire risk analysis.

Keywords: Data mining, clustering, high dimensions projected clustering, pruning.

Full Text:



  • There are currently no refbacks.

© 2017 International Journal of Global Research in Computer Science (JGRCS)
Copyright Agreement & Authorship Responsibility