期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2020
卷号：98
期号：15
页码：2977-2989
出版社：Journal of Theoretical and Applied
摘要：This research is motivated by the abundance of time series data stack found, often regarded as garbage and neglected due to the inability to find knowledge or interesting patterns from the data pile. Time series is one of the topics that is often associated with forecasting through a series of data that depends on time periods. The basic problem in time series data mining is how to present the knowledge contained therein, then how to find the rules of periodic data series and how to optimize the decision of the resulting time series data so that it can be used to predict in the future. Based on previous papers, there is no model to present knowledge in the form of rules in time series. In this paper the proposed model is RBT (Rule Best Time Series). The main process in RBT is to discretize periodic series to form sub-sequences, then these sub-sequences are grouped through measures of similarity with distance using euclidean, then the discovery of rules is applied to obtain hidden rules on temporal patterns and to rank with J-measures. From the results of this study time series data can be optimized, new knowledge or trends and patterns in time series databases that are uncertain and previously unknown can be generated. The decision or information can be used to display decisions, or forecasting in the future with an accuracy rate of the model mean absolute deviation (MAD) of 73%, forecasting accuracy of the mean squared deviation (MSE) of 87% and the percentage of the mean absolute percentage error of the MAPE of 4,7%.
关键词：Forecasting Time Series Data Mining;Rule Best Time Series
