Research On Spatiotemporal Data-oriented Mining Algorithms

With the rapid development of society,the amount of data generated by various sectors is increasing day by day.It becomes particularly important to make use of data quickly and efficiently to discover the value.As a kind of data mining branch,efficient mining have been applied in many areas.However,the large data with a large amount,many types,low value density,high efficiency,which requires the algorithm has a high space-time efficiency.This paper propose HUIMR algorithm on discovering high utility itemset(HUI).The algorithm is based on the MapReduce framework.It can accommodate big data environments.The HUIMR algorithm consists of counting and mining two stages.For the counting stage,MapReduce is used to calculate high transaction-weighted utilization items.While during the mining stage,high transaction-weighted utilization itemset tree is defined and HUIs is parallel mined by using MapReduce based on the pattern growth strategy.Based on the utility value and the historical data of the existing tags,we propose a utility-based parallel random forest algorithm.Random forest is made of several decision trees,thus the parallel random forest algorithm mainly consists of two processes:parallelization establishment and call of decision trees.And get final result by summarizing situation of each decision tree.Proved by experiment,the algorithm works well in processing large scale datasets.This paper studies a Visual Traffic Forecast System based on discovering high utility itemset.The system takes the traffic data of the intersection as input,preprocessing by aggregation and error correction.Then passes the data into the distributed file system,and mining the high utility itemset by HUIMR.It make predictions based on the itemset,and make historical traffic data visualization.