Wyniki wyszukiwania - BazTech

1

Ensembles of instance selection methods: A comparative study

Blachnik Marcin

International Journal of Applied Mathematics and Computer Science

|

2019

|

Vol. 29, no. 1

151--168

EN

Instance selection is often performed as one of the preprocessing methods which, along with feature selection, allows a significant reduction in computational complexity and an increase in prediction accuracy. So far, only few authors have considered ensembles of instance selection methods, while the ensembles of final predictive models attract many researchers. To bridge that gap, in this paper we compare four ensembles adapted to instance selection: Bagging, Feature Bagging, AdaBoost and Additive Noise. The last one is introduced for the first time in this paper. The study is based on empirical comparison performed on 43 datasets and 9 base instance selection methods. The experiments are divided into three scenarios. In the first one, evaluated on a single dataset, we demonstrate the influence of the ensembles on the compression–accuracy relation, in the second scenario the goal is to achieve the highest prediction accuracy, and in the third one both accuracy and the level of dataset compression constitute a multi-objective criterion. The obtained results indicate that ensembles of instance selection improve the base instance selection algorithms except for unstable methods such as CNN and IB3, which is achieved at the expense of compression. In the comparison, Bagging and AdaBoost lead in most of the scenarios. In the experiments we evaluate three classifiers: 1NN, kNN and SVM. We also note a deterioration in prediction accuracy for robust classifiers (kNN and SVM) trained on data filtered by any instance selection methods (including the ensembles) when compared with the results obtained when the entire training set was used to train these classifiers.

2

A local model and calibration set ensemble strategy for open-path FTIR gas measurement with varying temperature

Cięszczyk S.

Metrology and Measurement Systems

|

2013

|

Vol. 20, nr 3

513--524

EN

Open-Path Fourier Transform Infrared OP-FTIR spectrometers are commonly used for the measurement of atmospheric pollutants and of gases in industrial processes. Spectral interpretation for the determination of gas concentrations is based on the HITRAN database line-by-line modeling method. This article describes algorithms used to model gas spectra and to determine gas concentration under variable temperatures. Integration of individual rotational lines has been used to reduce the impact of spectrometer functions on the comparison of both measured and synthetic modeled spectra. Carbon monoxide was used as an example. A new algorithm for gas concentration retrieval consisting of two ensemble methods is proposed. The first method uses an ensemble of local models based on linear and non-linear PLS (partial least square) regression algorithms, while the second is an ensemble of a calibration set built for different temperatures. It is possible to combine these methods to decrease the number of regression models in the first ensemble. These individual models are appropriate for specific measurement conditions specified by the ensemble of the calibration set. Model selection is based on comparison of gas spectra with values determined from each local model.

3

Multi-label classification using error correcting output codes

Kajdanowicz T., Kazienko P.

International Journal of Applied Mathematics and Computer Science

|

2012

|

Vol. 22, no. 4

829-840

EN

A framework for multi-label classification extended by Error Correcting Output Codes (ECOCs) is introduced and empirically examined in the article. The solution assumes the base multi-label classifiers to be a noisy channel and applies ECOCs in order to recover the classification errors made by individual classifiers. The framework was examined through exhaustive studies over combinations of three distinct classification algorithms and four ECOC methods employed in the multi-label classification problem. The experimental results revealed that (i) the Bode-Chaudhuri-Hocquenghem (BCH) code matched with any multi-label classifier results in better classification quality; (ii) the accuracy of the binary relevance classification method strongly depends on the coding scheme; (iii) the label power-set and the RAkEL classifier consume the same time for computation irrespective of the coding utilized; (iv) in general, they are not suitable for ECOCs because they are not capable to benefit from ECOC correcting abilities; (v) the all-pairs code combined with binary relevance is not suitable for datasets with larger label sets.

4

Noise Detection for Latent Component Classification in Ensemble Method

Szupiluk R.

Przegląd Elektrotechniczny

|

2010

|

R. 86, nr 1

144-146

EN

We present a novel concept of the random noise detection applied in model integration process. The ensemble method is based on decomposition of the multivariate variable containing model results. The decomposition originating from Blind Signal Separation gives us the latent components with destructive or constructive impact on the prediction quality. The identification and elimination of the noisy signals should improve final prediction result. The validity of our concept is presented on the real problem of load forecasting in the Polish Power System.

PL

W artykule przedstawiono nową metodę detekcji szumów losowych zastosowana w procesie agregacji modeli. W rozwijanej metodzie agregacji zbieramy rezultaty poszczególnych modeli predykcyjnych w jednej wielowymiarowej zmiennej. Zakładamy, że zawiera ona konstruktywne oraz destrukcyjne dla wyników prognozy ukryte komponenty. Komponenty te możemy estymować metodami ślepej separacji sygnałów. Poprawna identyfikacji oraz eliminacja komponentów szumowych prowadzi do poprawy ostatecznych wyników predykcji. Potwierdzeniem skuteczności proponowanych rozwiązań jest przykład predykcji obciążenia systemu elektroenergetycznego.

5

ICA method and RBF neural network for prediction improvement

Szupiluk R., Siwek K., Wojewnik P., Ząbkowski T.

Przegląd Elektrotechniczny

|

2007

|

R. 83, nr 11

57-59

EN

In this paper we present a novel method for integration the prediction results by finding common latent components via independent component analysis. The latent components can have constructive or destructive influence on particular prediction results. After the elimination of the deconstructive signals we rebuilt the improved predictions using RBF neural networks. We check the method validity on the electricity load prediction task.

PL

W artykule przedstawiono nową metodę pozwalającą na łączenie wyników predykcji poprzez poszukiwanie ukrytych wspólnych składowych przy zastosowaniu procedury analizy składowych niezależnych. Składowe ukryte mogą mieć pozytywny lub negatywny wpływ na wyniki predykcji. Po wyeliminowaniu składowych niekorzystnych poprawiono model wykorzystując sztuczną sieć neuronową z radialną funkcją bazową. Poprawność metody sprawdzono na przykładzie predykcji poboru mocy.

6

Weighted ensemble boosting for robust activity recognition in video

Ivanov Y., Hamid R.

Machine Graphics and Vision

|

2006

|

Vol. 15, No. 3/4

415-427

EN

In this paper we introduce a novel approach to classifier combination, which we term Weighted Ensemble Boosting. We apply the proposed algorithm to the problem of activity recognition in video, and compare its performance to different classifier combination methods. These include Approximate Bayesian Combination, Boosting, Feature Stacking, and the more traditional Sum and Product rules. Our proposed Weighted Ensemble Boosting algorithm combines the Bayesian averaging strategy with the boosting framework, finding useful conjunctive feature combinations and achieving a lower error rate than the traditional boosting algorithm. The method demonstrates a comparable level of stability with respect to the classifier selection pool. We show the performance of our technique for a set of 6 types of classifiers in an office setting, detecting 7 classes of typical office activities.