A spatio-temporal pattern prediction is to forecast the future events based on the time and location information. The spatio-temporal pattern prediction has been done using various techniques but still not efficient for accurate prediction with minimum time while handling the big dataset since the dataset comprises the more features and more data. Therefore, the Sampling Distributive Chi-square based Stochastic Discriminant Random Decision Tree Classification (SDC-SDRDTC) technique is introduced to enhance the spatio-temporal pattern prediction accuracy (PA) with lesser time. Initially, Multi-dimensional Isomap Scaling based CURE data clustering algorithm finds all possible distributions (i.e., clusters) based on time and location from large dataset for pattern discovery. The CURE data clustering algorithm extracts the spatio-temporal pattern. After pattern extraction, Sampling Distribution Chi-squared Test is used for pattern selection from extracted pattern for prediction. In addition, Sampling Distributive Chi-Squared Test applied for selecting the spatio-temporal pattern based on the score value. After pattern selection, Stochastic Discriminant Random Decision Tree Classifier (SDRDTC) presents ensemble learning method for classification and prediction through constructing the multitude of decision trees. A classifier comprises several weak learners. For each decision tree, selected pattern are given as input randomly. The results of decision tree are combined and the votes are generated. The majority votes of patterns are considered as global spatio temporal pattern and it is used for prediction of future events with higher PA. Experimental analysis is performed with El Nino Dataset and taxi trajectory dataset with different metrics namely PA, false positive rate (FPR) and prediction time (PT). From the results, it is evident that SDC-SDRDTC technique obtains higher PA with lesser time and FPR than the conventional methods.
Volume 12 | 03-Special Issue
Pages: 1429-1440
DOI: 10.5373/JARDCS/V12SP3/20201395