Theoretical Method and Data Validation of Time Series Residual Analysis Based on Iterative Rnn Model
0
Citation
25
Reference
10
Related Paper
Abstract:
Currently, it is commonly believed that the residuals, also known as white noise, do not contain useful information, so that further analyses on them are often considered unnecessary. However, this paper raises doubts about this notion and believes that the residuals still contain valuable information, which traditional methods find difficult to get extracted. Therefore, to address this issue, based on the efficient market hypothesis, this paper first established an iterative RNN model, and used the model to make analyses on the residuals, in which, Bitcoin prices were selected as the research sample. Second, this paper managed to make predictions on subsequent Bitcoin prices, and make comparison between the predicted ones and the actual ones. It was eventually found that the iterative RNN model succeeded analyzing the residuals, with the prediction accuracy improved significantly. Subsequently, this paper conducted the robustness testing, which indicates that the iterative RNN model was suitable not only for cryptocurrency markets but also for stock markets. In short, this research indicates that the iterative RNN model surpasses traditional prediction models in terms of information absorption and analyses, and the approach used in the research is not only applicable to iterative RNN model but also to other similar models, such as iterative LSTM models, iterative CNN models, etc.This paper presents a novel time-series pattern and rule mining technique.The technique is that the time-series data waiting for mining is first converted into sub-time-series data,and then the knowledge underlying in the sub-time-series data is used as a guide to mine the original time series and extract the association rules from them.The paper gives the algorithm for mining time-series pattern or rules and illustrates the technique to be effective and feasible.
Cite
Citations (3)
Considering the researching background, this paper discussed the associated research of data mining technology, time series and financial time series data mining, separately. As following, the basic theory and methods of phase space reconstruction were analyzed in details. All of these provided the theoretical basis and technical feasibility to time series data mining based on phase space reconstruction. After contrasting the different means of time series pattern mining, we pointed out the problem of time series data mining framework TSDM, and presented the temporal patterns mining method based wave cluster systematically.
Cite
Citations (8)
Cite
Citations (14)
A time-series data mining technique is introduced in this paper. The mining technique of basing on varying-series is first to change the time-series data into its varying-series data, and then to use the information hiding in the latest time subseries of the time-series varying series to guide us to forecast the variety of the original time series. The mining approach of basing on rough set theory is to import the varying- series data set into a decision table first, and then to use the knowledge, underlying in the condition attribute set equivalence class, to estimate our forecasting object. Finally, the mining algorithm is given, and a concrete mining case is given to test the feasibility and the performance of the mining approach.
Decision table
Data set
Equivalence class (music)
Cite
Citations (1)
Feed forward
Cite
Citations (30)
A combination of embedding theorem and artificial intelligence along with residual analysis is used to analyze and forecast chaotic time series. Based on embedding theorem, the time series is reconstructed into proper phase space points and fed into a neural network whose weights and biases are improved using genetic algorithms. As the residuals of predicted time series demonstrated chaotic behavior, they are reconstructed as a new chaotic time series. A new neural network is trained to forecast future values of residual time series. The residual analysis is repeated several times. Finally, a neural network is trained to capture the relationship among the predicted value of the original time series, residuals, and the original time series. The method is applied to two chaotic time series, Mackey-Glass and Lorenz, for validation, and it is concluded that the proposed method can forecast the chaotic time series more effectively and accurately than existing methods.
Cite
Citations (30)
In this paper, the conceptions of the time-series varying series and the latest time subseries are presented, and the time-series mining technique based on the varying series and the time-series mining technique based on the rough set approach are proposed.The mining technique basing on varying series is first to convert the time series waiting for our studying into its varying series, and then to use the information hiding in the latest time subseries of the time-series varying series to guide us to forecast the variety of the original time series, the time-series mining technique based on the rough set approach is first to import the varying series data set into a decision table, and then to use the knowledge, which is underlying in the condition attribute set equivalence class in which the latest time subseries object is included, to estimate our forecasting object. Finally, the mining algorithm is given, and a concrete mining case is given to test the feasibility and the performance of the mining approach.
Cite
Citations (0)
According to the time series,the research and analysis of time-series association rules mining has been discussed in this paper,and a new algorithm for time-series association rules data mining based on sliding window and time-series tree with special structure has been proposed.By using this algorithm,the mining to frequent time-series which exceeds a given support count threshold has been acted to provide the decision support and trend prediction for users.The experiment results prove the validity and practicability of this algorithm.
Sliding window protocol
Cite
Citations (0)
Cite
Citations (3)
The traditional linear time series prediction algorithms for time series require high linearity,and nonlinear methods are generally modeling complex and have a large computation.For the above,this paper proposed an algorithm for time series prediction which based on trends point state model.The algorithm didn't regard to whether the time series forecast significant linear features,first digged out the similar sequence on time series through the coupling between sequences,and identified the corresponding trend points of similar sequence,then established the trend point state model and calculated the predicted value.Using this algorithm modeling is simple,and the complexity is low.Through simulation,the results show that the algorithm has a high prediction accuracy,especially for periodic time series.
Sequence (biology)
Cite
Citations (0)