Abstract: This paper develops an unscented grid-based filter
and a smoother for accurate nonlinear modeling and analysis
of time series. The filter uses unscented deterministic sampling
during both the time and measurement updating phases, to approximate
directly the distributions of the latent state variable. A
complementary grid smoother is also made to enable computing
of the likelihood. This helps us to formulate an expectation
maximisation algorithm for maximum likelihood estimation of
the state noise and the observation noise. Empirical investigations
show that the proposed unscented grid filter/smoother compares
favourably to other similar filters on nonlinear estimation tasks.
Abstract: By taking advantage of both k-NN which is highly
accurate and K-means cluster which is able to reduce the time of classification, we can introduce Cluster-k-Nearest Neighbor as "variable k"-NN dealing with the centroid or mean point of all subclasses generated by clustering algorithm. In general the algorithm of K-means cluster is not stable, in term of accuracy, for that reason we develop another algorithm for clustering our space which gives a higher accuracy than K-means cluster, less
subclass number, stability and bounded time of classification with respect to the variable data size. We find between 96% and 99.7 % of accuracy in the lassification of 6 different types of Time series by using K-means cluster algorithm and we find 99.7% by using the new clustering algorithm.
Abstract: The urban centers within northeastern Brazil are
mainly influenced by the intense rainfalls, which can occur after long
periods of drought, when flood events can be observed during such
events. Thus, this paper aims to study the rainfall frequencies in such
region through the wavelet transform. An application of wavelet
analysis is done with long time series of the total monthly rainfall
amount at the capital cities of northeastern Brazil. The main
frequency components in the time series are studied by the global
wavelet spectrum and the modulation in separated periodicity bands
were done in order to extract additional information, e.g., the 8 and
16 months band was examined by an average of all scales, giving a
measure of the average annual variance versus time, where the
periods with low or high variance could be identified. The important
increases were identified in the average variance for some periods,
e.g. 1947 to 1952 at Teresina city, which can be considered as high
wet periods. Although, the precipitation in those sites showed similar
global wavelet spectra, the wavelet spectra revealed particular
features. This study can be considered an important tool for time
series analysis, which can help the studies concerning flood control,
mainly when they are applied together with rainfall-runoff
simulations.
Abstract: The authors have been developing several models
based on artificial neural networks, linear regression models, Box-
Jenkins methodology and ARIMA models to predict the time series
of tourism. The time series consist in the “Monthly Number of Guest
Nights in the Hotels" of one region. Several comparisons between the
different type models have been experimented as well as the features
used at the entrance of the models. The Artificial Neural Network
(ANN) models have always had their performance at the top of the
best models. Usually the feed-forward architecture was used due to
their huge application and results. In this paper the author made a
comparison between different architectures of the ANNs using
simply the same input. Therefore, the traditional feed-forward
architecture, the cascade forwards, a recurrent Elman architecture and
a radial based architecture were discussed and compared based on the
task of predicting the mentioned time series.
Abstract: Many digital signal processing, techniques have been used to automatically distinguish protein coding regions (exons) from non-coding regions (introns) in DNA sequences. In this work, we have characterized these sequences according to their nonlinear dynamical features such as moment invariants, correlation dimension, and largest Lyapunov exponent estimates. We have applied our model to a number of real sequences encoded into a time series using EIIP sequence indicators. In order to discriminate between coding and non coding DNA regions, the phase space trajectory was first reconstructed for coding and non-coding regions. Nonlinear dynamical features are extracted from those regions and used to investigate a difference between them. Our results indicate that the nonlinear dynamical characteristics have yielded significant differences between coding (CR) and non-coding regions (NCR) in DNA sequences. Finally, the classifier is tested on real genes where coding and non-coding regions are well known.
Abstract: ANNARIMA that combines both autoregressive integrated moving average (ARIMA) model and artificial neural network (ANN) model is a valuable tool for modeling and forecasting nonlinear time series, yet the over-fitting problem is more likely to occur in neural network models. This paper provides a hybrid methodology that combines both radial basis function (RBF) neural network and auto regression (AR) model based on binomial smoothing (BS) technique which is efficient in data processing, which is called BSRBFAR. This method is examined by using the data of Canadian Lynx data. Empirical results indicate that the over-fitting problem can be eased using RBF neural network based on binomial smoothing which is called BS-RBF, and the hybrid model–BS-RBFAR can be an effective way to improve forecasting accuracy achieved by BSRBF used separately.
Abstract: Many systems in the natural world exhibit chaos or non-linear behavior, the complexity of which is so great that they appear to be random. Identification of chaos in experimental data is essential for characterizing the system and for analyzing the predictability of the data under analysis. The Lyapunov exponents provide a quantitative measure of the sensitivity to initial conditions and are the most useful dynamical diagnostic for chaotic systems. However, it is difficult to accurately estimate the Lyapunov exponents of chaotic signals which are corrupted by a random noise. In this work, a method for estimation of Lyapunov exponents from noisy time series using unscented transformation is proposed. The proposed methodology was validated using time series obtained from known chaotic maps. In this paper, the objective of the work, the proposed methodology and validation results are discussed in detail.
Abstract: An adaptive software reliability prediction model
using evolutionary connectionist approach based on Recurrent Radial
Basis Function architecture is proposed. Based on the currently
available software failure time data, Fuzzy Min-Max algorithm is
used to globally optimize the number of the k Gaussian nodes. The
corresponding optimized neural network architecture is iteratively
and dynamically reconfigured in real-time as new actual failure time
data arrives. The performance of our proposed approach has been
tested using sixteen real-time software failure data. Numerical results
show that our proposed approach is robust across different software
projects, and has a better performance with respect to next-steppredictability
compared to existing neural network model for failure
time prediction.
Abstract: This paper presents a procedure for estimating VAR
using Sequential Discounting VAR (SDVAR) algorithm for online
model learning to detect fraudulent acts using the telecommunications
call detailed records (CDR). The volatility of the VAR is observed
allowing for non-linearity, outliers and change points based on the
works of [1]. This paper extends their procedure from univariate
to multivariate time series. A simulation and a case study for
detecting telecommunications fraud using CDR illustrate the use of
the algorithm in the bivariate setting.
Abstract: Much time series data is generally from continuous dynamic system. Firstly, this paper studies the detection of the nonlinearity of time series from continuous dynamics systems by applying the Phase-randomized surrogate algorithm. Then, the Delay Vector Variance (DVV) method is introduced into nonlinearity test. The results show that under the different sampling conditions, the opposite detection of nonlinearity is obtained via using traditional test statistics methods, which include the third-order autocovariance and the asymmetry due to time reversal. Whereas the DVV method can perform well on determining nonlinear of Lorenz signal. It indicates that the proposed method can describe the continuous dynamics signal effectively.
Abstract: A concern that researchers usually face in different
applications of Artificial Neural Network (ANN) is determination of
the size of effective domain in time series. In this paper, trial and
error method was used on groundwater depth time series to determine
the size of effective domain in the series in an observation well in
Union County, New Jersey, U.S. different domains of 20, 40, 60, 80,
100, and 120 preceding day were examined and the 80 days was
considered as effective length of the domain. Data sets in different
domains were fed to a Feed Forward Back Propagation ANN with
one hidden layer and the groundwater depths were forecasted. Root
Mean Square Error (RMSE) and the correlation factor (R2) of
estimated and observed groundwater depths for all domains were
determined. In general, groundwater depth forecast improved, as
evidenced by lower RMSEs and higher R2s, when the domain length
increased from 20 to 120. However, 80 days was selected as the
effective domain because the improvement was less than 1% beyond
that. Forecasted ground water depths utilizing measured daily data
(set #1) and data averaged over the effective domain (set #2) were
compared. It was postulated that more accurate nature of measured
daily data was the reason for a better forecast with lower RMSE
(0.1027 m compared to 0.255 m) in set #1. However, the size of input
data in this set was 80 times the size of input data in set #2; a factor
that may increase the computational effort unpredictably. It was
concluded that 80 daily data may be successfully utilized to lower the
size of input data sets considerably, while maintaining the effective
information in the data set.
Abstract: Power Spectral Density (PSD) of quasi-stationary processes can be efficiently estimated using the short time Fourier series (STFT). In this paper, an algorithm has been proposed that computes the PSD of quasi-stationary process efficiently using offline autoregressive model order estimation algorithm, recursive parameter estimation technique and modified sliding window discrete Fourier Transform algorithm. The main difference in this algorithm and STFT is that the sliding window (SW) and window for spectral estimation (WSA) are separately defined. WSA is updated and its PSD is computed only when change in statistics is detected in the SW. The computational complexity of the proposed algorithm is found to be lesser than that for standard STFT technique.
Abstract: This paper proposes a method that discovers time series event patterns from textual data with time information. The patterns are composed of sequences of events and each event is extracted from the textual data, where an event is characteristic content included in the textual data such as a company name, an action, and an impression of a customer. The method introduces 7 types of time constraints based on the analysis of the textual data. The method also evaluates these constraints when the frequency of a time series event pattern is calculated. We can flexibly define the time constraints for interesting combinations of events and can discover valid time series event patterns which satisfy these conditions. The paper applies the method to daily business reports collected by a sales force automation system and verifies its effectiveness through numerical experiments.
Abstract: In this paper we introduce a novel method for
the characterization of synchronziation and coupling effects
in multivariate time series that can be used for the analysis
of EEG or ECoG signals recorded during epileptic seizures.
The method allows to visualize the spatio-temporal evolution
of synchronization and coupling effects that are characteristic
for epileptic seizures. Similar to other methods proposed for
this purpose our method is based on a regression analysis.
However, a more general definition of the regression together
with an effective channel selection procedure allows to use the
method even for time series that are highly correlated, which
is commonly the case in EEG/ECoG recordings with large
numbers of electrodes. The method was experimentally tested
on ECoG recordings of epileptic seizures from patients with
temporal lobe epilepsies. A comparision with the results from
a independent visual inspection by clinical experts showed
an excellent agreement with the patterns obtained with the
proposed method.
Abstract: Recognizing behavioral patterns of financial markets
is essential for traders. Japanese candlestick chart is a common tool to
visualize and analyze such patterns in an economic time series. Since
the world was introduced to Japanese candlestick charting, traders
saw how combining this tool with intelligent technical approaches
creates a powerful formula for the savvy investors.
This paper propose a generalization to box counting method of
Grassberger-Procaccia, which is based on computing the correlation
dimension of Japanese candlesticks instead commonly used 'close'
points. The results of this method applied on several foreign
exchange rates vs. IRR (Iranian Rial). Satisfactorily show lower
chaotic dimension of Japanese candlesticks series than regular
Grassberger-Procaccia method applied merely on close points of
these same candles. This means there is some valuable information
inside candlesticks.
Abstract: The focus of this paper is to construct daily time series
exchange rate forecast models of Samoan Tala/USD and Tala/AUD
during the year 2008 to 2012 with neural network The performance
of the models was measured by using varies error functions such as
Root Square mean error (RSME), Mean absolute error (MAE), and
Mean absolute percentage error (MAPE). Our empirical findings
suggest that AR (1) model is an effective tool to forecast the
Tala/USD and Tala/AUD.
Abstract: Since the pioneering work of Zadeh, fuzzy set theory has been applied to a myriad of areas. Song and Chissom introduced the concept of fuzzy time series and applied some methods to the enrollments of the University of Alabama. In recent years, a number of techniques have been proposed for forecasting based on fuzzy set theory methods. These methods have either used enrollment numbers or differences of enrollments as the universe of discourse. We propose using the year to year percentage change as the universe of discourse. In this communication, the approach of Jilani, Burney, and Ardil is modified by using the year to year percentage change as the universe of discourse. We use enrollment figures for the University of Alabama to illustrate our proposed method. The proposed method results in better forecasting accuracy than existing models.
Abstract: In this paper, the application of multiple Elman neural networks to time series data regression problems is studied. An ensemble of Elman networks is formed by boosting to enhance the performance of the individual networks. A modified version of the AdaBoost algorithm is employed to integrate the predictions from multiple networks. Two benchmark time series data sets, i.e., the Sunspot and Box-Jenkins gas furnace problems, are used to assess the effectiveness of the proposed system. The simulation results reveal that an ensemble of boosted Elman networks can achieve a higher degree of generalization as well as performance than that of the individual networks. The results are compared with those from other learning systems, and implications of the performance are discussed.
Abstract: The prediction of financial time series is a very
complicated process. If the efficient market hypothesis holds, then the predictability of most financial time series would be a rather
controversial issue, due to the fact that the current price contains already all available information in the market. This paper extends
the Adaptive Neuro Fuzzy Inference System for High Frequency
Trading which is an expert system that is capable of using fuzzy reasoning combined with the pattern recognition capability of neural networks to be used in financial forecasting and trading in high
frequency. However, in order to eliminate unnecessary input in the
training phase a new event based volatility model was proposed.
Taking volatility and the scaling laws of financial time series into consideration has brought about the development of the Intraday Seasonality Observation Model. This new model allows the observation of specific events and seasonalities in data and subsequently removes any unnecessary data. This new event based
volatility model provides the ANFIS system with more accurate input
and has increased the overall performance of the system.
Abstract: This paper deals with the application of a well-known neural network technique, multilayer back-propagation (BP) neural network, in financial data mining. A modified neural network forecasting model is presented, and an intelligent mining system is developed. The system can forecast the buying and selling signs according to the prediction of future trends to stock market, and provide decision-making for stock investors. The simulation result of seven years to Shanghai Composite Index shows that the return achieved by this mining system is about three times as large as that achieved by the buy and hold strategy, so it is advantageous to apply neural networks to forecast financial time series, the different investors could benefit from it.