Classification of Time-Interval and Hybrid Sequential Temporal Patterns

摘要

Due to the rapid growth of information systems that manage temporal data, efficient and automated classification techniques are of great importance. For instance, timely and accessible temporal data enhances critical financial operations such as predicting future stock prices. Similarly, in medical domain, classifying temporal data, which is relevant to patients or critical operations, leads to efficient control and recovery from severe problems. Therefore, time is an essential dimension to many domain-specific problems. This research introduces Temporal-ROLEX; a framework to categorize temporal data that effectively induces semantic temporal patterns. This paper presents an efficient rule-based classification approach for categorizing temporal data. The contributions of this research are 1) formulating Semantic Temporal patterns as a basic classification features, and 2) introducing an induction technique to discriminate semantic temporal patterns. The proposed framework extends ROLEX-SP approach to handle the classification of temporal data in different domains. To illustrate the design, the article provides a detailed mathematical description that relies on set-theory to model the framework of Temporal-ROLEX. Furthermore, this paper provides a detailed description of proposed algorithms to facilitate implementing and reproducing the results. To evaluate the effectiveness of the Temporal-ROLEX, we performed extensive experiments on a weather temporal dataset. Also, the F-measure and support values on weather dataset are reported as well as a scalability and sensitivity analysis to assess the capability of Temporal-ROLEX to work with temporal datasets. Findings indicate a significant improvement of Temporal-ROLEX over some existing techniques. Specifically, Temporal-ROLEX achieves significant enhancement using sequential temporal pattern over existing state-of-the-art techniques. On the other hand, Temporal-ROLEX achieves average performance using hybrid temporal patterns. Finally, the results have been analyzed and justified the factors that affect the performance in both cases.