Find the Best Rule of Time Series Data Mining with Cluster Analysis

作     者：Relita Buaton Muhammad Zarlis Herman Mawengkang Syahril Efendi 

作者机构：Graduate Program Of Computer Science Department Of Computer Science Faculty of Computer Science and Information Technology Universitas Sumatera Utara Medan Indonesia 

出 版 物：《Journal of Physics: Conference Series》 

年 卷 期：2021年第1830卷第1期

学科分类：07[理学] 0702[理学-物理学] 

摘      要：Data mining is an analytical process of knowledge discovery in large and complex data sets. Many studies wish to explore data, to find information so that knowledge can be obtained through the grouping process, classification, rules discovery, associations and data mining visualization which shows similarity. Periodic data often occurs in business applications and sciences that has big size, high dimension and continuously updated. The similarity in periodic data is based on several approaches. One of common approaches is to transform periodic series into other domains so that dimensions are reduced, followed by index mechanism. Many studies of time series do not give optimal result because limited to extracting data not able to represent time series and its pattern which is then change into rules. Rules can be found in time series data, but they are still constrained by over fitting and difficult to present. It causes time series data and non linier function of data mining decision can t be optimal. The basic idea in the method proposed is to do periodic discretization for sub-sequential formation. These sub-sequences are grouped through a measure of similarity. The simple rule-finding technique is applied to obtain hidden rules in the temporal pattern. The optimal time series data expected to generate the uncertainty trend, previously unknown and can be used to make decisions or forecasting in the future.