Machine Learning Approaches for Time-Series Data Based on Self-Organizing Incremental Neural Network

Abstract

In this paper, we introduce machine learning algorithms of time-series data based on Self-organizing Incremental Neural Network (SOINN). SOINN is known as a powerful tool for incremental unsupervised clustering. Using a similarity threshold based and a local error-based insertion criterion, it is able to grow incrementally and to accommodate input patterns of on-line non-stationary data distribution. These advantages of SOINN are available for modeling of time-series data. Firstly, we explain an on-line supervised learning approach, SOINN-DTW, for recognition of time-series data that are based on Dynamic TimeWarping (DTW). Second, we explain an incremental clustering approach, Hidden-Markov-Model Based SOINN (HBSOINN), for time-series data. This paper summarizes SOINN based time-series modeling approaches (SOINN-DTW, HBSOINN) and the advantage of SOINN-based time-series modeling approaches compared to traditional approaches such as HMM.