Stochastic correlation coefficient ensembles for variable selection

A new framework for interval wavelength selection based on wavelength importance clustering

BACKGROUND

Wavelength selection is one of the key steps in spectral analysis and plays an irreplaceable role in improving model prediction accuracy and computational efficiency. High-dimensional spectral datasets contain substantial irrelevant information and redundant variables. Whereas, at current stage, such problem can be solved by existing abundant wavelength selection methods. However, it is difficult to achieve the balance between strong wavelength interpretability and prediction accuracy by those methods. As a result, there is an urgent need for a new method that can reach the point of balance.

RESULTS

we propose a new framework for wavelength selection based on wavelength importance clustering (WIC) which attempts to establish a hierarchical relationship between wavelength points and attributions of response through a clustering algorithm, consequently, performing combinatorial and filtering to obtain the optimal wavelength combinations. In this paper, a new wavelength selection method (WIC-WRCKF) is constructed based on WIC, and four commonly used wavelength selection methods are selected to be compared with WIC-WRCKF. A large number of experiments are carried out on three publicly available datasets as well, namely, wheat, corn, and tablets. Compared with other methods, WIC-WRCKF has the highest prediction accuracy with high stability on the three datasets, and the number of wavelengths selected is small and highly interpretative, indicating that WIC-WRCKF has a better predictive ability.

SIGNIFICANCE

The wavelength selection method can significantly improve the model prediction accuracy, and the WIC architecture can effectively exploit the essence of the spectral data, which has great potential in the application of wavelength selection.

Ensembling Variable Selectors by Stability Selection for the Cox Model

As a pivotal tool to build interpretive models, variable selection plays an increasingly important role in high-dimensional data analysis. In recent years, variable selection ensembles (VSEs) have gained much interest due to their many advantages. Stability selection (Meinshausen and Bühlmann, 2010), a VSE technique based on subsampling in combination with a base algorithm like lasso, is an effective method to control false discovery rate (FDR) and to improve selection accuracy in linear regression models. By adopting lasso as a base learner, we attempt to extend stability selection to handle variable selection problems in a Cox model. According to our experience, it is crucial to set the regularization region Λ in lasso and the parameter λ_min properly so that stability selection can work well. To the best of our knowledge, however, there is no literature addressing this problem in an explicit way. Therefore, we first provide a detailed procedure to specify Λ and λ_min. Then, some simulated and real-world data with various censoring rates are used to examine how well stability selection performs. It is also compared with several other variable selection approaches. Experimental results demonstrate that it achieves better or competitive performance in comparison with several other popular techniques.