Search Results for « Time series prediction »

A novel methodology is proposed for clustering multivariate time series data using energy distance defined in Székely and Rizzo (2013). Specifically, a dissimilarity matrix is formed using the energy distance statistic to measure separation between the finite dimensional distributions for the component time series. Once the pairwise dissimilarity matrix is calculated, a hierarchical clustering method is then applied to obtain the dendrogram. This procedure is completely nonparametric as the dissimilarities between stationary distributions are directly calculated without making any model assumptions. In order to justify this procedure, asymptotic properties of the energy distance estimates are derived for general stationary and ergodic time series. The method is illustrated in a simulation study for various component time series that are either linear or nonlinear. Finally the methodology is applied to two examples; one involves GDP of selected countries and the other is population size of various states in the U.S.A. in the years 1900 -1999.

Conformal prediction provides machine learning models with prediction sets that offer theoretical guarantees, but the underlying assumption of exchangeability limits its applicability to time series data. Furthermore, existing approaches struggle to handle multi-step ahead prediction tasks, where uncertainty estimates across multiple future time points are crucial. We propose JANET (Joint Adaptive predictioN-region Estimation for Time-series), a novel framework for constructing conformal prediction regions that are valid for both univariate and multivariate time series. JANET generalises the inductive conformal framework and efficiently produces joint prediction regions with controlled K-familywise error rates, enabling flexible adaptation to specific application needs. Our empirical evaluation demonstrates JANET's superior performance in multi-step prediction tasks across diverse time series datasets, highlighting its potential for reliable and interpretable uncertainty quantification in sequential data.

Recurrent neural networks and exceedingly Long short-term memory (LSTM) have been investigated intensively in recent years due to their ability to model and predict nonlinear time-variant system dynamics. The present paper delivers a comprehensive overview of existing LSTM cell derivatives and network architectures for time series prediction. A categorization in LSTM with optimized cell state representations and LSTM with interacting cell states is proposed. The investigated approaches are evaluated against defined requirements being relevant for an accurate time series prediction. These include short-term and long-term memory behavior, the ability for multimodal and multi-step ahead predictions and the according error propagation. Sequence-to-sequence networks with partially conditioning outperform the other approaches, such as bidirectional or associative networks, and are best suited to fulfill the requirements.

The space-time foliation Sigma compatible with the gravitational field g on a 4-manifold M determines a fibration pi of M, pi : M -> N is a surjective submersion over the 1-dimensional leaves space N. M is then written as a disjoint union of the leaves of Sigma, which are 3-dimensional spacelike surfaces on M. The decomposition, TM=Sigma + T^0 M, also implies that we can define a lift of the curves on N to curves (non-spacelike) on M. The stable causality condition M coincides with Sigma being a causal space-time distribution, generated by an exact timelike 1-form omega^0 = dt where t is some real function on M. In this case M is written as a disjoint union of a family of spacelike 3-surfaces of constant t, which cover D^+(S) of a initial 3-surface S of M.

Recent studies have suggested frequency-domain Data augmentation (DA) is effec tive for time series prediction. Existing frequency-domain augmentations disturb the original data with various full-spectrum noises, leading to excess domain gap between augmented and original data. Although impressive performance has been achieved in certain cases, frequency-domain DA has yet to be generalized to time series prediction datasets. In this paper, we found that frequency-domain augmentations can be significantly improved by two modifications that limit the perturbations. First, we found that limiting the perturbation to only dominant frequencies significantly outperforms full-spectrum perturbations. Dominant fre quencies represent the main periodicity and trends of the signal and are more important than other frequencies. Second, we found that simply shuffling the dominant frequency components is superior over sophisticated designed random perturbations. Shuffle rearranges the original components (magnitudes and phases) and limits the external noise. With these two modifications, we proposed dominant shuffle, a simple yet effective data augmentation for time series prediction. Our method is very simple yet powerful and can be implemented with just a few lines of code. Extensive experiments with eight datasets and six popular time series models demonstrate that our method consistently improves the baseline performance under various settings and significantly outperforms other DA methods. Code can be accessed at https://kaizhao.net/time-series.

Long Short Term Memory (LSTM) is among the most popular deep learning models used today. It is also being applied to time series prediction which is a particularly hard problem to solve due to the presence of long term trend, seasonal and cyclical fluctuations and random noise. The performance of LSTM is highly dependent on choice of several hyper-parameters which need to be chosen very carefully, in order to get good results. Being a relatively new model, there are no established guidelines for configuring LSTM. In this paper this research gap was addressed. A dataset was created from the Indian stock market and an LSTM model was developed for it. It was then optimized by comparing stateless and stateful models and by tuning for the number of hidden layers.

We develop a general framework for constructing distribution-free prediction intervals for time series. Theoretically, we establish explicit bounds on conditional and marginal coverage gaps of estimated prediction intervals, which asymptotically converge to zero under additional assumptions. We obtain similar bounds on the size of set differences between oracle and estimated prediction intervals. Methodologically, we introduce a computationally efficient algorithm called \texttt{EnbPI} that wraps around ensemble predictors, which is closely related to conformal prediction (CP) but does not require data exchangeability. \texttt{EnbPI} avoids data-splitting and is computationally efficient by avoiding retraining and thus scalable to sequentially producing prediction intervals. We perform extensive simulation and real-data analyses to demonstrate its effectiveness compared with existing methods. We also discuss the extension of \texttt{EnbPI} on various other applications.