©2010 International Journal of Computer Applications (0975 - 8887) Volume 1 – No. 23 53 Continuous K-means computation over moving objects by designing different threshold dissemination protocols CH.Dayakar Reddy Dr A Govardhan Talari Swapna A Brahmananda Reddy MCA,M.Phil,(P.hD) B.Tech,M.Tech,Ph.D B.Tech,M.Tech B.Tech,M.Tech,(Ph.D) Associate Professor Professor & Principal Assistant Professor Assistant Professor C M R CET JNTUCEH,Jagtiyal TRR Engg. College C M R CET Hyderabad Hyderabad Hyderabad Hyderabad ABSTRACT In this paper, we study designing different threshold protocols that are used to monitor a set of moving objects in k-means computation at a server. In a given data set P, a k-means query returns k points in space, such that average squared distance between each point in p and its nearest center is minimized. Reevaluating k-means every time there is an object update imposes heavy burden on the server and the clients where it reduces the computation and communication costs. The proposed method assigns each moving object a threshold and uses multiple servers for monitoring locations of distinct set of objects and their updates when it crosses the range boundary. Index Terms—k-Means, continuous monitoring, query processing. 1.Intoduction The k-means computation is crucial in many practical applications, facility location planning, spatial decision making and clustering. Our work focuses on continuous k-means monitoring over moving objects, which has numerous practical applications. For instance, real-time traffic control systems. A simple method for continuous k-means monitoring, hereafter called REF (short for reference solution), works as follows: When the system starts at time τ= 0, every object reports its location, and the server computes the k-means set M(0) through the HC(hill climbing) algorithm. Subsequently, (τ > 0),whenever an object moves, it sends a location update. The server obtains M(τ) by executing HC on the updated locations, using M(τ - 1) as the seeds. The rationale is that M( τ )is expected to be more similar to M( τ -1) than a random seed set, reducing the number of HC iterations.REF produces high-quality results because it continuously follows every object update. On the other hand, it incurs large communication and computation cost due to the frequent updates and recomputations.To eliminate these problems, we propose a threshold-based k- means monitoring (TKM) method, based on the framework of Fig. 1. In addition to k, a continuous k-means query specifies a tolerance ∆. The computation of M(0) is the same as in REF. Our contributions are as follows: 1. We present TKM, a general framework for continuous k-means monitoring over moving objects. 2. We propose HC*, an improved HC, which minimizes the cost of each iteration by only considering a small subset of the objects. 3. We model the threshold assignment task as a constrained optimization problem, and derive the corresponding mathematical formulas. .4. We design different mechanisms for the dissemination of thresholds, depending on the computational capabilities of the objects. 2 k-Means Computation for Static Data Most data mining literature has applied HC for solving k-means. Fig. 2 shows a general version of the algorithm.