Machine Learning Applications in Real-World Time Series Problems

Nowadays, information systems produce a vast amount of data, which is expected to register an exponential growth in the next few years. This data is commonly treated as a time series, which are chronologically collected data representing a function that varies over time. Time series are presented in a wide range of science fields, from hydrology or palaeoclimatology to economy, among others. This chapter highlights some of the primary time series data mining (TSDM) tasks, among which time series preprocessing, segmentation or prediction are widespread in the literature. Machine learning (ML) techniques, along with other time series approaches, are detailed throughout a set of real-world applications, such as the wave height time series reconstruction, the detection and prediction of tipping points in palaeoclimatology time series and the forecasting of low-visibility events produced by the existence of fog or convective situations, among others. Finally, this chapter also summarizes the principal contributions of the authors to these problems.