Fuzzy-clustering time series: Population-based an enhanced technique

• Published in: 2017 Artificial Intelligence and Robotics (IRANOPEN) pp. 84 - 90
• Main Authors: Mehrmolaei, Soheila, Keyvanpour, Mohammad Reza
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.04.2017
• Subjects: Algorithm design and analysis; Clustering algorithms; Data mining; fuzzy-clustering; Indexes; Optimization; Particle swarm optimization; population-based technique; time series; Time series analysis
• DOI: 10.1109/RIOS.2017.7956448

One of the main challenging subjects of data mining is fuzzy-clustering time series in real-world applications. Its reason can be time-series data characteristics that include high dimensional, large volume and existence of temporal ordering in data. So far, many studies have performed about issues such as addressing time-series data high dimension and applying a different effect of each dimension in clustering results. In many of the recent published resources, feature-weighting process in time-series data has employed as one effective way to overcome mentioned problems. Appropriate weight's assignment can discuss as one challenge in mentioned resources. Hence, aim of this paper is optimization of the feature-weighting process in time series clustering task. For this aim, a technique is proposed based on the combination of two fundamental concepts, including features effect-based weighting and optimization in order to optimize weight's assignment in time series clustering. Proposed technique is made possibility use of the merits of two concepts in order to improving performance fuzzy-clustering task in time-series data. Fuzzy Particle Swarm Optimization (FPSO) is used as a population-based technique for optimization of feature-weighting process and enhancing performance of clustering. Proposed technique is evaluated based on three indexes validity that includes Partition Coefficient (PC), Partition Entropy (PE) and Davios-Bouldin (DB). Experimental results are indicated that our proposed technique is efficient and can represent encouraging results. Also, clustering results by population-based a proposed technique is demonstrated the improvement of performance of fuzzy-clustering time series than other conventional techniques.