SDR: A Novel Similarity Measure Using Curve Fitting Method for Time Series Data Clustering
Huahui Yang, Meng Chen, Cheng Wang, Yunzhi Yao · 2019
Research in similarity measures for time series clustering and classification has produced promising results since the emerge of data mining and artificial intelligence. However, the significant scale and phase variances of sequences within the same cluster, constitute challenges calling for the quest of correct partitions. In this research, we are trying to address the problem using the Shape-Distance Ratio (SDR) similarity measure (SIM). SDR adopts curve fitting and it applies a segmented method for short-length sequences which are divided into several segments using sliding windows. The optimal similar segments between curve and objective sequence are found in the windows. This measure was used to process raw data. We have compared the SDR similarity measure with other state-of-the-art distance measures. The experimental results clearly show that SDR offers an effective SIM for time series data. Then a hand-craft is presented to evaluate the performance of our proposed method in clustering task.