nTreeClus: A model - basd partitioning clustering of sequential data

Abstract

The overwhelming presence of categorical/sequential data in diverse domains emphasizes the importance of sequence mining. The challenging nature of sequences proves the need for a continuing research to find a more accurate and faster approach providing a better understanding of their (dis)similarities. The thesis proposes a new model-based approach for clustering sequence data, namely nTreeClus. Proposed ap proach is at the intersection of the conceptions behind existing approaches, including Decision Tree Learning, k-mers, and autoregressive models for categorical time series, culminating with a novel numerical representation of the categorical sequences. This new representation is used to perform clustering considering the inherent patterns in categorical time series. Furthermore, the only parameter of the method, the window size (n), is examined and the robustness of the method to its parameter has been shown. Under different simulated scenarios and using various internal and external cluster validation indices, the performance of the method is analyzed and reported. Finally, empirical evaluation using synthetic and real datasets, protein sequences and categorical time series, declares that nTreeClus is competitive or superior to most of the state-of-the-art methods.