Combining Machine Learning and Symbolic Representation of Time Series for Classification of Behavioural Patterns

The emergence of affordable wireless sensors has enabled the development of information systems combining sophisticated data processing and machine learning algorithms for pattern recognition. In many cases, these systems deal with time-series data, continuously gathered by sensors that compile detailed activity records. However, these datasets are frequently affected by numerous problems, including noisy data acquisition, missing data and utilization of inefficient techniques for information representation, which lead to deficient performance in machine learning applications. In this paper, we introduce a novel method to combine the efficient symbolic representation of time-series data with machine learning to improve the performance of classification systems tailored to detection of behavioural patterns of interest.