A new distance in pattern clustering on longitudinal data
Yi Liu, Nianlong Luo · 2014
Clustering as an unsupervised learning method is still an effective way for pattern analysis on longitudinal data. Because of the characteristics of pattern clustering on longitudinal data, accumulated minor noise and data shifting, the traditional distance for clustering algorithm based on partitioning, such as Euclidean distance, could not perform very well. A new distance for partitioning clustering algorithm, Max-Difference distance, is proposed to solve these problems which could not be solved by Euclidean distance. According to the result of three experiments, Max-Difference shows its effectiveness for longitudinal data and proves that it can work well for pattern clustering on longitudinal data.