Computational Approaches for Time Series Analysis and Prediction. Data-Driven Methods for Pseudo-Periodical Sequences.

摘要

Time series data mining is one branch of data mining. Time series analysisudand prediction have always played an important role in human activities andudnatural sciences. A Pseudo-Periodical time series has a complex structure,udwith fluctuations and frequencies of the times series changing over time. Currently,udPseudo-Periodicity of time series brings new properties and challengesudto time series analysis and prediction.udThis thesis proposes two original computational approaches for time seriesudanalysis and prediction: Moving Average of nth-order Difference (MANoD)udand Series Features Extraction (SFE). Based on data-driven methods, theudtwo original approaches open new insights in time series analysis and predictionudcontributing with new feature detection techniques. The proposedudalgorithms can reveal hidden patterns based on the characteristics of timeudseries, and they can be applied for predicting forthcoming events.udThis thesis also presents the evaluation results of proposed algorithms onudvarious pseudo-periodical time series, and compares the predicting resultsudwith classical time series prediction methods. The results of the originaludapproaches applied to real world and synthetic time series are very good andudshow that the contributions open promising research directions.