A methodology for training set instance selection using mutual information in time series prediction

Stojanović, Miloš B.; Božić, Miroslav; Stanković, Milena; Stajić, Zoran

doi:10.1016/j.neucom.2014.03.006

Cited by 47 publications

(15 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Classical whitebox methods for forecasting demand include linear regression [15] and ARIMA [16] models. As mentioned in [17] and [18] these classical methods provide poor estimates if the consumption behavior is nonlinear, nonstationary, and not known prior to model construction. As a result black-box machine learning algorithms such as ANNs and support vector machines (SVMs) are of growing interest for forecasting the power demand of consumers [19].…”

Section: Background and Motivationmentioning

confidence: 99%

Nonparametric Demand Forecasting and Detection of Energy Aware Consumers

Hoiles

Krishnamurthy

2015

IEEE Trans. Smart Grid

View full text Add to dashboard Cite

To increase the reliability of the power grid and reduce the risk of power supply failure, demand-side management (DSM) is of central importance. In this paper, a nonparametric test is applied to detect if the demand behavior of consumers is consistent with time-of-day electricity tariff initiatives. The test is based on Afriat's theorem in economics and has the unique feature that it provides necessary and sufficient conditions to detect if the price-demand behavior is consistent with utility maximization (i.e., the test detects demand-responsive consumers) without prior knowledge of the consumer's utility function. For consumers that are responsive to time-of-day pricing initiatives, a nonparametric learning algorithm is used to forecast power demands for unobserved electricity tariffs. The nonparametric learning algorithm can be used in anticipatory control structures in a DSM framework to achieve power usage objectives. Real-world data from Ontario's power system and numerical examples illustrate the accuracy of the nonparametric test and nonparametric learning algorithm for forecasting consumer demand. IndexTerms-Afriat's theorem, artificial neural network (ANN), demand-side management (DSM), revealed preferences, smart grid, utility maximization.

show abstract

Section: Background and Motivationmentioning

confidence: 99%

Nonparametric Demand Forecasting and Detection of Energy Aware Consumers

Hoiles

Krishnamurthy

2015

IEEE Trans. Smart Grid

View full text Add to dashboard Cite

show abstract

“…However, their usage is commonly limited to functions of one or two variables because the number of samples needed for PDF estimation increases exponentially with the number of variables [19]. In this study, k-nearest neighbor-based MI estimation method proposed by Kraskov et al [20] is used.…”

Section: Estimation Of the Mutual Information Using Knearest Neighborsmentioning

confidence: 99%

“…The basic idea is to estimate I(X, Y) from the distances in spaces X, Y and Z from z i to its k-nearest neighbors, averaged over all z i [19]. Let us define…”

Section: Estimation Of the Mutual Information Using Knearest Neighborsmentioning

confidence: 99%

Integration of mutual information and CRPSO-based fuzzy model for stock index forecasting

Yu¹,

Kim²

2015

Proceedings of the 2015 Conference of the International Fuzzy Systems Association and the European Society for Fuzzy Logic and

View full text Add to dashboard Cite

In this paper, the integration of mutual information (MI) and fuzzy model is proposed to predict stock indexes with complex and non-linear characteristics. Technical indicators are considered as initial input candidates and significant inputs are determined by MI-based input selection method. To identify the structures and parameters of fuzzy models simultaneously, cooperative random learning particle swarm optimization (CRPSO), proposed by Zhao et al., is used. To confirm the effectiveness, the proposed method and comparison methods are applied to the Korea Composite Stock Price Index (KOSPI). The experimental results show that the proposed method, on average, outperforms other comparison methods.

show abstract

“…It is a key problem in the field of pattern recognition and widely used for processing high-dimensional dataset. Compared to feature extraction, feature selection could maintain physical properties of the original features and has better interpretability, which has been widely used in time series analysis [12] and pattern classification [13]. For example, in order to improve the prediction accuracy and reduce training time, selecting proper training set instance is an important preprocessing step for long-term time series prediction.…”

Section: Introductionmentioning

confidence: 99%