Hybrid wavelet-neural network models for time series data

Download
2021-3-3
Kılıç, Deniz Kenan
The thesis aims to combine wavelet theory with nonlinear models, particularly neural networks, to find an appropriate time series model structure. Data like financial time series are nonstationary, noisy, and chaotic. Therefore using wavelet analysis helps better modeling in the sense of both frequency and time. S&P500 (∧GSPC) and NASDAQ (∧ IXIC) data are divided into several components by using multiresolution analysis (MRA). Subsequently, each part is modeled by using a suitable neural network structure. In this step, the design of the model is formed according to the pattern of the subseries. Then predictions of each subseries are combined. The combined prediction result is compared to the original time series’s prediction result using only a nonlinear model. Moreover, wavelets are used as an activation function for LSTM networks to form a hybrid LSTM-Wavenet model. Furthermore, the hybrid LSTM-Wavenet model is fused with MRA as a proposed method. In brief, it is studied whether using MRA and hybrid LSTM-Wavenet model decreases the loss or not for both S&P500 and NASDAQ data. Four different modeling methods are used: LSTM, LSTM+MRA, hybrid LSTM-Wavenet, hybrid LSTMWavenet+MRA (the proposed method). Results show that using MRA and wavelets as an activation function together decreases error values the most.

Suggestions

Multiresolution analysis of S&P500 time series
KILIC, Deniz Kenan; Uğur, Ömür (2018-01-01)
Time series analysis is an essential research area for those who are dealing with scientific and engineering problems. The main objective, in general, is to understand the underlying characteristics of selected time series by using the time as well as the frequency domain analysis. Then one can make a prediction for desired system to forecast ahead from the past observations. Time series modeling, frequency domain and some other descriptive statistical data analyses are the primary subjects of this study: i...
Semi-Bayesian Inference of Time Series Chain Graphical Models in Biological Networks
Farnoudkia, Hajar; Purutçuoğlu Gazi, Vilda (null; 2018-09-20)
The construction of biological networks via time-course datasets can be performed both deterministic models such as ordinary differential equations and stochastic models such as diffusion approximation. Between these two branches, the former has wider application since more data can be available. In this study, we particularly deal with the probabilistic approaches for the steady-state or deterministic description of the biological systems when the systems are observed though time. Hence, we consider time s...
A temporal neuro-fuzzy approach for time-series analysis
Yılmaz (Şişman), Nuran Arzu; Alpaslan, Ferda Nur; Department of Computer Engineering (2003)
The subject of this thesis is to develop a temporal neuro-fuzzy system for fore- casting the future behavior of a multivariate time series data. The system has two components combined by means of a system interface. First, a rule extraction method is designed which is named Fuzzy MAR (Multivari- ate Auto-regression). The method produces the temporal relationships between each of the variables and past values of all variables in the multivariate time series system in the form of fuzzy rules. These rules may ...
Loop-based conic multivariate adaptive regression splines is a novel method for advanced construction of complex biological networks
Ayyıldız Demirci, Ezgi; Purutçuoğlu Gazi, Vilda; Weber, Gerhard Wilhelm (2018-11-01)
The Gaussian Graphical Model (GGM) and its Bayesian alternative, called, the Gaussian copula graphical model (GCGM) are two widely used approaches to construct the undirected networks of biological systems. They define the interactions between species by using the conditional dependencies of the multivariate normality assumption. However, when the system's dimension is high, the performance of the model becomes computationally demanding, and, particularly, the accuracy of GGM decreases when the observations...
Prediction Model Selection with Frequency Check on Decomposed Time Series
Büyükşahin, Ümit Çavuş; Ertekin Bolelli, Şeyda (2019-08-22)
High prediction accuracies at time series modeling and forecasting is of the utmost importance for a variety of application domains. Various time series prediction methods exist that use linear and nonlinear models separately or combination of both. These methods highly increase prediction performance results when they are applied on a many number of stationary components obtained by more sophisticated decomposition techniques. Although these stationary components are easily predictable, they each have diff...
Citation Formats
D. K. Kılıç, “Hybrid wavelet-neural network models for time series data,” Ph.D. - Doctoral Program, Middle East Technical University, 2021.