Ryszard Szupiluk

Model Improvement by the Statistical Decomposition

In this paper we propose applying multidimensional decompositions for modeling improvement. Results generated by different models usually include both wanted and destructive components. Many of the components are common to all the models. Our aim is to find the basis variables with the positive and the negative influence on the modeling task. It will be perofrmed with multidimensional transforamtions such as ICA and PCA.

Prediction improvement via smooth component analysis and neural network mixing

In this paper we derive a novel smooth component analysis algorithm applied for prediction improvement. When many prediction models are tested we can treat their results as multivariate variable with the latent components having constructive or destructive impact on prediction results. The filtration of those destructive components and proper mixing of those constructive should improve final prediction results. The filtration process can be performed by neural networks with initial weights computed from smooth component analysis. The validity and high performance of our concept is presented on the real problem of energy load prediction.

Noise detection for ensemble methods

In this paper we present a novel noisy signal identification method applied in ensemble methods for destructive components classification. Typically two main signal properties like variability and predictability are described by the same second order statistic characteristic. In our approach we postulate to separate measure of the signal internal dependencies and their variability. The validity of the approach is confirmed by the experiment with energy load data.

Smooth component analysis as ensemble method for prediction improvement

In this paper we apply a novel smooth component analysis algorithm as ensemble method for prediction improvement. When many prediction models are tested we can treat their results as multivariate variable with the latent components having constructive or destructive impact on prediction results. We show that elimination of those destructive components and proper mixing of those constructive can improve the final prediction results. The validity and high performance of our concept is presented on the problem of energy load prediction.

The noise identification method based on divergence analysis in ensemble methods context

In this paper we propose a divergence based method for noise detection in ensemble method context where the prediction results from different models are treated as a multidimensional variable that contains constructive and destructive latent components. The crucial stage is the proper destructive and constructive components classification. We propose to calculate the noisiness of the particular latent component as the divergence from chosen reference noise. It allows us to identify the wide range of noises besides the typical signals with close analytical form such as Gaussian or uniform. The real data experiment with load energy prediction confirms presented methodology.

Blind Signal Separation Methods for Integration of Neural Networks Results

In this paper it is proposed to apply blind signal separation methods to improve a neural network prediction. Results generated by any regression model usually include both constructive and destructive components. In case of a few models, some of the components can be common to all of them. Our aim is to find the basis elements and distinguish the components with the constructive influence on the modelling quality from the destructive ones. After rejecting the destructive elements from the models results it is observed the enhancement of the results in terms of some standard error criteria. The validity and high performance of the concept is presented on the real problem of energy load prediction

Grade analysis for energy usage patterns segmentation based on smart meter data

The Grade Correspondence Analysis (GCA) with posterior clustering and visualization is introduced and applied to smart meter data on individual household level. The main task of this analysis is to reveal the latent structure of electricity usage patterns and to propose a two dimensional segmentation taking into account the usage of selected home appliances and time of their usage. This provides the solutions applicable in smart metering systems that can support usage forecasting and contribute to higher energy awareness.

Multiplicative ICA algorithm for interaction analysis in financial markets

In this article we present a new method for the analysis of dependencies in case of multivariate time series. In this approach, we assume that the set of time series representing the various financial instruments creates a multidimensional variable. Such a multidimensional variable is decomposed into independent components which enable to analyze the morphology of given financial instruments and to identify the hidden interdependencies. We propose a new multiplicative version of the Natural Gradient ICA algorithm that could be used in automated trading systems or modeling environments. The presented method is tested on real stock markets data.

Multistage covariance approach to measure the randomness in financial time series analysis

The paper presents a new method for randomness assessment in data with temporal structure. In this approach we perform multistage covariance analysis on several parts of the signal to synthesize information about variability and internal dependencies included in its structure. This allows us to identify deterministic cycles or to detect the level of randomness in signals what is an important issue for the design of transactional, prediction and filtration systems. To confirm validity of the proposed method we tested it on simulated and real financial time series.

Independent Component Analysis for Filtration in Data Mining

In this paper we propose a new method for data mining prediction improvement. There are many prediction models with different advantages. Each model brings some positive as well as some negative features in terms of prediction quality. Different criteria can indicate different models as the best solution. Our aim is to utilize results from many models, identify common destructive components as precisely as possible and eliminate them. This will be done by Independent Component Analysis (ICA). The modified ICA -algorithm for effective problem solving will be proposed.