Presentation on theme: "Moving Pattern Detection in Spatio Temporal Data Mining BY SEKHAR NALLURI VENGALA RAO PACHVA SURYA KOGANTI."— Presentation transcript:
Moving Pattern Detection in Spatio Temporal Data Mining BY SEKHAR NALLURI VENGALA RAO PACHVA SURYA KOGANTI
PROBLEM DEFINITION INCREASE IN USE OF WIRELESS COMMUNICATION DEVICES LEADS TO THE DEVELOPMENT OF LOCATION BASED SERVICES. THE OBJECTIVE IS TO TRACE THE MOVING OBJECTS WITH LOW RESPONSE TIME AND MINIMUM UTILIZATION OF RESOURCES TO DELIVER EFFICIENT LOCATION SERVICES.
Paper 1 A CLUSTERING BASED APPROACH FOR DISCOVERING INTERESTING PLACES IN A SINGLE TRAJECTORY
Motivation With the development of many location sensors, a lot of trajectories of users and moving objects can be identified This creates an appropriate basis for developing efficient new methods for mining moving objects Semantic clustering of trajectories left behind moving objects is an important aspect in spatio temporal data mining
Introduction Spatio temporal data is growing substantially, with the increasing use of wireless communication devices An algorithm CB-SMoT (Cluster Based Stops and Moves of Trajectories) has been provided to extract stops and moves from trajectory sample points. Based on traditional algorithm DBSCAN EPS parameter in the algorithm CB-SMoT can dramatically affect the quality of clustering A new method of calculating the EPS value is proposed
Cont.. Moving patterns are extracted from stops and moves To model stops and moves, the user has to specify the places of interest, since stops and moves are specified in advance from an application point of view The main drawback of this assumption is that important places that may lead to the discovery of interesting patterns can be missed if they are not known by the user. The proposed algorithm can extract some unknown stops
Cont’d.. The distances between two neighbor points in the trajectory are not distributed by Gaussian curve as shown in figure 1 and figure 2 So more appropriate method is proposed to calculate the parameter Eps
Definitions Trajectory Sample A trajectory sample is a list of space-time points: Stops represent the important places of a trajectory where the moving object has stayed for a minimal amount of time
Definitions Move: A move of a trajectory T with respect to an application A is: a maximal contiguous sub-trajectory of T in between two temporally consecutive stops of T ; OR a maximal contiguous sub-trajectory of T in between the starting point of T and the first stop of T ; OR a maximal contiguous sub-trajectory of T in between the last stop of T and the last point of T ; OR the trajectory itself, if T has no stops
CB-SMoT Algorithm Well known density based clustering algorithm The CB-SMoT is interested specially in discovering clusters in a single trajectory and also considers time Two major steps in CB-SMoT In first step, the slower parts of a trajectory called potential stops are identified using the variation of the DBSCAN algorithm In the second step, the algorithm identifies where these potential stops found in the first step are located
Cont’d.. The Eps parameter indicates the absolute distance used to calculate the neighborhood of a point A trajectory T can be viewed as a list of distances di between two consecutive points pi and pi+1. These distances have an arithmetic mean μ and a standard deviation. With these two parameters it is possible to plot the appropriate Gaussian curve. The Quantile function is defined as:
New method to calculate EPS According to the distances histogram, it is easy to distinguish the fast speed part and the slow speed part. Then we calculate the arithmetic mean μ and the standard deviation. To denote (μ1, ı1) with respect to the distances di between two consecutive points pi and pi +1 in the whole trajectory and (μ2,ı2) with respect to the distances di between two consecutive points pi and pi+1 in the slow part.
Cont’d According to the trajectory plotted, we can get (μ1, ı1) = (1.537, 0.845) and (μ2, ı2) = (0.441, 0.095). Authors thought that the distance di is subjected to Gaussian Curve so, these situations can be standardized and can utilize the normal distribution table to reference the E values WRT different mean and SD.
Conclusion Spatio temporal Data Mining is still infancy What kinds of patterns can be extracted from trajectories? Which methods and algorithms should be applied to extract them? Trajectory clustering is only a little attempt. Future Work Future work is needed for finding the better method to calculate the parameters Eps and MinTime for the purpose of improvement the quality of clustering
Paper 2 Efficient STMPM(Spatio-Temporal Moving Pattern Mining) Using Moving Sequence Tree
Introduction The problem with various pattern mining methods already available are: Increasing Execution time Increasing Size of memory for search A new algorithm is proposed to overcome these problems and to efficiently extract the periodical or sequential frequent moving patterns.
Definition Spatio temporal Frequent moving pattern mining Given moving object database MD, user specified minimum support factor min_sup, and the constraint of time interval between spatial scopes max_gap, spatio-temporal frequent moving pattern mining searches all frequent moving sequences that satisfies the minimum support factor.
TERMS Min_sup: Lowest value of support factor that has to be satisfied to determine the sequence S is frequent or not Max_gap : The maximum time gap tj-tj-1, moving time from a specific area to another closely located area
STMPM PROCEDURE THERE ARE TWO MAIN PROCESSES IN THE STMPM PROCEDURE PREPROCESSING: SPATIO-TEMPORAL ATTRIBUTES OF MOVING OBJECT ARE TRANSFORMED INTO APPROPRIATE FORM FOR PATTERN MINING. PATTERN MINING: MOVING SEQUENCE TREE IS CREATED TO SEARCH FREQUENT PATTERNS THAT SATISFY MINIMUM SUPPORT FACTOR.
Contd… A MOVING SEQUENCE IS A SEQUENTIAL LIST. IT NEEDS TO SATISFY THE CONSTRAINT OF TIME INTERVAL AMONG LOCATIONS THAT FORM A SEQUENCE. CONTAINS OPERATION GENERALIZES LOCATION ATTRIBUTES OF THE OBJECT TO SPATIAL SCOPE. DURING OPERATION GENERALIZES TEMPORAL ATTRIBUTES TO TEMPORAL SCOPE. GENERALIZED DATA IS SUMMARIZED INTO ONE SEQUENCE ITEM IF SPATIAL AND TEMPORAL ATTRIBUTES ARE IDENTICAL.
PATTERN MINING A MOVING SEQUENCE TREE IS A HASH TREE-BASED SEQUENCE TREE FORMED BY THE SEQUENCE OF EACH OBJECT. EXTRACTS FREQUENT PATTERN THAT SATISFIES MIN_SUP.
Set of generalized moving sequences
MINING PROCESS STEP 1 : SEARCH FOR SET OF FREQUENT 1-SEQUENCES AND TRANSFORMATION OF TRANSACTION DATA WITH MIN_SUP HIGHER THAN 2.
Cont’d.. STEP2 : CONSTRUCTION OF MOVING SEQUENCE TREES ONE MOVING SEQUENCE CREATES 2N-1 PARTIAL SEQUENCES WHEN THERE ARE N SEQUENCE ITEMS, EXCLUDING A NULL SET.
CHARECTERISTICS LIMITED DATASET: EXTRACTS THE SET OF HISTORICAL DATA OF A MOVING OBJECT. SEQEXTRACTOR : CREATES SEQUENTIAL LIST OF EACH OBJECT. CONTAINS : IT IS A SPATIAL OPERATION TO GENERALIZE LOCATION ATTRIBUTE OF A MOVING OBJECT. DURING: IT IS A TIME INTERVAL OPERATION TO GENERALIZE TEMPORAL ATTRIBUTE OF A MOVING OBJECT IN THE SET OF MOVING SEQUENCES. FREQPATTERNEXTRACTOR : CREATES A MOVING SEQUENCE TREES AND EXTRACTS FREQUENT MOVING PATTERNS.
Experimental Results PERFORMANCE CRITERION IS EFFICIENCY IN EXECUTION TIME FOR PATTERN MINING, IT IS MEASURED BY USING THE MINIMUM SUPPORT FACTOR. GEOMETRY DATA USED IS ADMINISTRATIVE DISTRICT AND ROAD NETWORK DATA OF SEOUL. HISTORICAL DATA IS CREATED BY MEASURING THE DRIVING HISTORY OF TAXIS.
Contd.. CHARACTERISTICS OF EXPERIMENTAL DATA EXECUTION TIMES FOR EACH PATTERN MINING ARE MEASURED BY FIXING THE TIME RANGE TO 10 HOURS.
PAPER-3 AN ENERGY SAVING STRATEGY FOR OBJECT TRACKING IN SENSOR NETWORKS BY MINING SEAMLESS TEMPORAL MOVING PATTERNS
RELATED WORK HARDWARE DESIGN OPTIMIZATION PROBLEM OF THE COMMUNICATION COST BY INACTIVATING THE RF RADIOS OF IDLE SENSOR NODES SOFTWARE DESIGN APPROACH DEVELOPED SOME TREE STRUCTURES FOR EFFICIENT OBJECT TRACKING BY CONSIDERING THE PHYSICAL NETWORK STRUCTURE.
APPROACH ENERGY SAVING FOR TRACKING OBJECTS IN SENSOR NETWORKS IS DONE BY STMP-MINE. PREDICTION STRATEGIES EMPLOY THE DISCOVERED SEAMLESS TEMPORAL MOVEMENT PATTERNS TO REDUCE THE PREDICTION ERRORS FOR ENERGY SAVING. AVOIDS ENERGY EXPENSIVE COMPONENTS. THE PREDICTION-BASED STRATEGIES UTILIZING STMPS ARE PSTMP PES+PSTMP
Assumptions NETWORK MODEL : A SENSOR NODE IS ACTIVATED ONLY IF THERE IS OBJECT IN ITS COVERAGE/SENSING REGION. TRAJECTORY OF EACH OBJECT IS REPRESENTED IN THE FORM OF S =, WHERE LI REPRESENTS THE SENSOR NODE LOCATION AT TIME TI. STMP IN THE FORM AS P = IK SEMANTICALLY MEANS THE REPRESENTATIVE TIME INTERVAL BETWEEN TWO TRAVERSED LOCATIONS. SEAMLESS TEMPORAL MOVEMENT RULE (STMR) INCORPORATED INTO THE LOCATION PREDICTION MECHANISMS. RT = →
STMP – Mine Algorithm TIME INTERVAL AGGREGATION TABLES ARE OBTAINED BY MANIPULATE THE TEMPORAL INFORMATION OF THE MOVEMENT LOGS. CLUSTERING TECHNIQUE IS EMPLOYED TO ACHIEVE AGGREGATION.
Contd… FOR A DISCOVERED STMP, PT =, THE DEFINITIONS OF CONFIDENCE CONF(PT) AND STRENGTH(PT) ARE GIVEN AS:
Simulation Model POPULAR METRICS NAMED TOTAL ENERGY CONSUMED (TEC) IS EVALUATED FOR THE PROPOSED PREDICTION STRATEGIES UNDER DIFFERENT TIME CONSTRAINTS. 80% OF THE SIMULATED DATA ARE USED FOR TRAINING TO OBTAIN STMRS, AND THE REST 20% ARE TAKEN AS TESTING SET FOR OBJECT TRACKING THE NETWORK IS MODELLED AS A MESH NETWORK WITH SIZE |W| = 20*20 WITH OBJECTS
Contd… THE BEHAVIOR OF MOVING OBJECTS IN THE OTSNS IS EVENT DRIVEN. TWO PARAMETERS LE AND PE TO MODEL THE AVERAGE LENGTH AND THE EVENT PROBABILITY. THE LENGTH OF EACH EVENT IS MODELLED BY POISSON DISTRIBUTION THE EVENT PROBABILITY INDICATES THE PROBABILITY FOR AN OBJECT TO ADHERE TO A CERTAIN EVENT, AND IT IS MODELLED BY NORMAL DISTRIBUTION. THE SENSING COVERAGE RANGE IS 15M AND THE AVERAGE OBJECT VELOCITY IS SET AS 15 M/S
COMPARISION OF PREDICTION STRATEGIES
PSTMP PSTMP-N-gram and PSTMP- N+-gram in terms of TEC and missing rate with TOP-N varied from 1 to 7. It is observed that the average number of STMRs stored in each sensor node with length greater or equal to 2 is about 5.34 in average, which is much less than that with length equal to 1 (about 10.56).
CONCLUSION PAPER1 PROPOSES TRAJECTORY CLUSTERING METHOD TO CALCULATE THE EPS VALUE WHICH SIGNIFICANTLY IMPROVE THE QUALITY OF CLUSTERING. PAPER2 SUGGESTED A STMPM ALGORITHM USING MOVING SEQUENCE TREE THAT MINIMIZES THE TIME NECESSARY FOR MINING AND THE AMOUNT OF MEMORY REQUIRED, SO THAT PATTERN MINING CAN BE CARRIED OUT SMOOTHLY. PAPER3 PROPOSE A SEAMLESS DATA MINING ALGORITHM NAMED STMP-MINE WITHOUT DEFINING SEGMENTING TIME UNIT TO EFFICIENTLY DISCOVERING THE SEAMLESS TEMPORAL MOVEMENT PATTERNS (STMPS) OF OBJECTS IN SENSOR NETWORKS
PROS STMPM ALGORITHM MINIMIZES THE TIME NECESSARY FOR MINING AND THE AMOUNT OF MEMORY REQUIRED. NO NEED OF SEGMENTING TIME UNIT IN OTSN’S. VELOCITY OF OBJECT IS NOT NEEDED. LOW MISSING RATE OF THE OBJECT IS POSSIBLE. ABUNDANT MOVING PATTERNS CAN BE FOUND WITHOUT OMISSION OF SHORT TIME INTERVAL
CONS THE KINDS OF PATTERNS EXTRACTED FROM TRAJECTORIES ARE UNANSWERED IN CLUSTERING BASED APPROACH. A MINING METHOD THAT SEARCHES FOR THE REGULAR MOVEMENT OF A MOVING OBJECT IS NOT DISCUSSED. ONLY THE TEMPORAL MOVING PATTERNS IS CONSIDERED FOR OBJECT TRACKING IN OTSN’S
COMAPRISION AND ANALAYSIS PAPER1PAPER2PAPER3 Eps and Min time are calculated. Less execution time and minimum memory space Total energy consumed is very less Kinds of patterns from the trajectory are not discussed Pattern Mining using Moving sequence tree is shown Pattern mining is performed to track moving object Both spatio and temporal attributes are considered Only temporal moving pattern are considered Efficient clustering technology Uses min Support factor to frequent patterns Clustering is used rather than segmenting unit time.
FUTURE WORK FUTURE WORK IS NEEDED FOR FINDING THE BETTER METHOD TO CALCULATE THE PARAMETERS EPS AND MINTIME. PATTERN MINING TECHNIQUES THAT NOT ONLY HAVE LOCATION HISTORY OF A MOVING OBJECT BUT ALSO THE INFORMATION ABOUT THE BEHAVIORS LIKE VELOCITY, DIRECTION ETC NEEDS TO BE ACCOMPLISHED THE TIME THE OBJECT SPENT IN STAYING AT CERTAIN LOCATION IS TO BE TRACED OUT.
REFERENCES S. ELNEKAVE, M. LAST, O. MAIMON "INCREMENTAL CLUSTERING OF MOBILE OBJECTS",STDM07, IEEE, 2007 J. ALLEN, "MAINTAINING KNOWLEDGE ABOUT TEMPORAL INTERVALS", COMM. OF THE ACM, VOL.26, NO.11, S. Y. HAN, "SPATIO-TEMPORAL MOVING SEQUENCE PATTERN MINING", EWHA WOMANS UNIV., KOREA, MS THESIS, T. PALPANAS, A. MENDELZON, WEB PREFETCHING USING PARTIAL MATCH PREDICTION, IN: PROC. OF THE 4TH WEB CACHING WORKSHOP, R. AGRAWAL, R. SRIKANT, MINING SEQUENTIAL PATTERNS, IN: PROC. OF THE 11TH INT’L CONF. ON DATA ENGINEERING, 1995, PP